This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

In how many distinct ways can flocks be formed?
A problem in sheep combinatorics

Johanna Langner* and Henryk A. Witek
Abstract

In this short paper, we extend the concept of the strict order polynomial ΩP(n)\Omega_{P}^{\circ}(n), which enumerates the number of strict order-preserving maps ϕ:P𝒏\phi:P\rightarrow\boldsymbol{n} for a poset PP, to the extended strict order polynomial EP(n,z)\text{E}_{P}^{\circ}(n,z), which enumerates analogous maps for the elements of the power set 𝒫(P)\mathcal{P}(P). The problem at hand immediately reduces to the problem of enumeration of linear extensions for the subposets of PP. We show that for every QPQ\subset P a given linear extension vv of QQ can be associated with a unique linear extension ww of PP. The number of such linear extensions vv (of length kk) associated with a given linear extension ww of PP can be expressed compactly as (delP(w)k)\binom{\text{del}_{P}(w)}{k}, where delP(w)\text{del}_{P}(w) is the number of deletable elements of ww defined in the text. Consequently the extended strict order polynomial EP(n,z)\text{E}_{P}^{\circ}(n,z) can be represented as follows

EP(n,z)=w(P)k=0p(delP(w)pk)(n+des(w)k)zk.\text{E}_{P}^{\circ}(n,z)=\sum_{w\in\mathcal{L}(P)}\sum_{k=0}^{p}\binom{\text{del}_{P}(w)}{p-k}\binom{n+\text{des}(w)}{k}z^{k}.

The derived equation can be used for example for solving the following combinatorial problem: Consider a community of pp shepherds, some of whom are connected by a master-apprentice relation (expressed as a poset PP). Every morning, kk of the shepherds go out and each of them herds a flock of sheep. Community tradition stipulates that each of these kk shepherds will herd at least one and at most nn sheep, and an apprentice will always herd fewer sheep than his master (or his master’s master, etc). In how many ways can the flocks be formed? The strict order polynomial answers this question for the case in which all pp shepherds go to work, and the extended strict order polynomial considers also all the situations in which some of the shepherds decide to take a day off.

Department of Applied Chemistry and Institute of Molecular Science, National Chiao Tung University, University Rd., 30010 Hsinchu, Taiwan

* [email protected]

1 Notation and Definitions

1.1 Standard terminology

The current communication closely follows the poset terminology introduced in Stanley’s book [1]. The reader familiar with the terminology can jump directly to Subsection 1.2. A partially ordered set PP, or poset for short, is a set together with a binary relation <P<_{P}. In this manuscript, we are concerned with finite posets PP consisting of pp elements and with strict partial orders, meaning that the relation <P<_{P} is irreflexive, transitive and asymmetric. An induced subposet QPQ\subset P is a subset of PP together with the order <Q<_{Q} inherited from PP which is defined by s<Pts<Qts<_{P}t\iff s<_{Q}t. The symbol << shall denote the usual relation ,,larger than” in \mathbb{N}. The symbol [n][\,n\,] stands for the set {1,2,,n}\left\{1,2,\ldots,n\right\}, and (n,m)(n,m) stands for the set {n+1,n+2,,m1}\left\{n+1,n+2,\ldots,m-1\right\}. The symbol 𝒏\boldsymbol{n} represents the chain 1<2<3<<n1<2<3<...<n. A map ϕ:P\phi:P\rightarrow\mathbb{\mathbb{\mathbb{N}}} is a strict order-preserving map if it satisfies s<Ptϕ(s)<ϕ(t)s<_{P}t\Rightarrow\phi(s)<\phi(t). A natural labeling of a poset PP is an order-preserving bijection ω:P[p]\omega:P\rightarrow[\,p\,]. A linear extension of PP is an order-preserving bijection σ:P𝒑\sigma:P\rightarrow\boldsymbol{p}. A linear extension σ\sigma can be represented as a permutation ωσ1\omega\circ\sigma^{-1} expressed as a sequence w=w1w2wpw=w_{1}w_{2}\ldots w_{p} with wi=ω(σ(i)1)w_{i}=\omega(\sigma{}^{-1}(i)); the sequence ww shall also be referred to as a linear extension in the following. The set of all such sequences ww is denoted by (P)\mathcal{L}\left(P\right) and is referred to as the Jordan-Hölder set of PP. If two subsequent labels wiw_{i} and wi+1w_{i+1} in ww stand in the relation wi>wi+1w_{i}>w_{i+1}, then the index ii is called a descent of ww. The total number of descents of ww is denoted by des(w)\text{des}(w). The strict order polynomial ΩP(n)\Omega_{P}^{\circ}(n) of a poset PP [2, 3, 1], which enumerates the strict order-preserving maps ϕ:P[n]\phi:P\rightarrow[\,n\,], is given by

ΩP(n)=w(P)(n+des(w)p).\Omega_{P}^{\circ}(n)=\sum_{w\in\mathcal{L}(P)}\binom{n+\text{des}(w)}{p}. (1)

1.2 Non-standard terminology

We will often construct—by slight abuse of notation—a subposet of PP by specifying a set of labels D[p]D\subset[\,p\,]: The expression PDP\setminus D stands for the induced subposet with the elements {pP|ω(p)D}\left\{p\in P\,|\,\omega(p)\notin D\right\}. Clearly the subposet constructed in this way has p#Dp-\#D elements; and the full set 𝒫(P)\mathcal{P}(P) of subposets of PP stands in direct correspondence to the power set of [p][\,p\,]: 𝒫(P)={PD|D𝒫([p])}\mathcal{P}(P)=\left\{P\setminus D\,|\,D\in\mathcal{P}([\,p\,])\right\}. Similarly, if ww is a sequence in (P)\mathcal{L}\left(P\right) and DD is a subset of [p][\,p\,], let us denote by wDw\setminus D the subsequence obtained by deleting all the elements of DD from ww. For example, 13245{1,4}=32513245\setminus\{1,4\}=325. Clearly, deleting some arbitrary set DD from two different sequences may produce the same sequence: for example, 13245{1,4}=325=32154{1,4}13245\setminus\{1,4\}=325=32154\setminus\{1,4\}. We will later (Lemma 12) see that deleting deletable elements (see Def. 6) from two distinct sequences always results in two distinct subsequences.

Further, let us slightly change the representation of linear extensions of subposets: Normally, one would assign to each subposet Q=PDQ=P\setminus D a new natural labeling ωQ:Q[q]\omega^{Q}:Q\rightarrow[\,q\,], and then express the linear extensions of QQ as sequences of the elements of [q][\,q\,]. Instead, we avoid re-labeling each subposet, and use instead the labeling ω:Q[p]D\omega:Q\rightarrow[\,p\,]\setminus D inherited from PP. Then, a linear extension σ\sigma of QQ is represented by a sequence w=w1wqw=w_{1}\ldots w_{q} defined in the usual way: wi=ω(σ(i)1)w_{i}=\omega(\sigma{}^{-1}(i)). The set of such sequences shall still be denoted by (Q)\mathcal{L}(Q). Using this notation, it is now easy to see (but properly demonstrated later in Lemma 14) that if ww is a linear extension of PP, then wDw\setminus D is a linear extension of PDP\setminus D.

2 Main results

In this short communication, we extend the concept of the strict order polynomial ΩP(n)\Omega_{P}^{\circ}(n) given by Eq. (1) to the extended strict order polynomial EP(n,z)\text{E}_{P}^{\circ}(n,z) given by Eq. (2), which enumerates and classifies the totality of strict order-preserving maps ϕ:Q𝒏\phi:Q\rightarrow\boldsymbol{n} with QPQ\subset P. We show below in Theorem 3 that there exists a compact combinatorial expression characterizing EP(n,z)\text{\emph{E}}_{P}^{\circ}(n,z). In the following, we shall always assume that PP is a poset with with pp elements, a strict order <P<_{P} , and a natural labeling ω\omega. Subposets of PP are always assumed to be induced.

Definition 1.

The extended strict order polynomial EP(n,z)\text{E}_{P}^{\circ}(n,z) of a poset PP is defined as

EP(n,z)=QPΩQ(n)z#Q,\text{E}_{P}^{\circ}(n,z)=\sum_{Q\subset P}\Omega_{Q}^{\circ}(n)z^{\#Q}, (2)

where the sum runs over all the induced subposets QQ of PP.

Example 2.

Consider a family of three shepherds: Fiadh, Fiadh’s father Aidan, and Aidan’s father Lorcan. Every day, some of the shepherds go out and each herd a flock of at least one and at most nn sheep. Aidan always herds more sheep than Fiadh, and Lorcan always herds more sheep than both Fiadh and Aidan. How many possible ways are there of assigning flock sizes to the shepherds?

Refer to caption
Figure 1: The poset formed by the three shepherds, shown in (a), is isomorphic to the chain 𝟑\boldsymbol{3}. Situations such as the one depicted in (b), where all three shepherds herd a flock of sheep, are counted by the strict order polynomial ΩP(n)\Omega_{P}^{\circ}(n). The extended strict order polynomial EP(n,z)\text{E}_{P}^{\circ}(n,z) also counts situations such as the one shown in (c), where only a subset of the shepherds are present.

The three shepherds together with the seniority relation form a poset PP isomorphic to the chain 𝟑\boldsymbol{3}: Fiadh<PAidan<PLorcan\text{Fiadh}<_{P}\text{Aidan}<_{P}\text{Lorcan}, see Fig. 1 (a). Let us denote the number of sheep in Fiadh’s flock by n1n_{1}, the size of Aidan’s flock by n2n_{2} and the size of Lorcan’s flock by n3n_{3}; then the above conditions tell us that 1n1<n2<n3n1\leq n_{1}<n_{2}<n_{3}\leq n. On a day when all three shepherds go to work, such as depicted in Fig. 1 (b), the numbers n1n_{1}, n2n_{2} and n3n_{3} can be chosen in ΩP(n)=(n3)\Omega_{P}^{\circ}(n)=\binom{n}{3} ways. When only Fiadh and Aidan go to work, see Fig. 1 (c), i.e. for the subposet Q={Fiadh,Aidan}Q=\{\text{Fiadh},\text{Aidan}\}, we have to choose the two numbers n1n_{1} and n2n_{2} such that 1n1<n2n1\leq n_{1}<n_{2}\leq n; there are ΩQ(n)=(n2)\Omega_{Q}^{\circ}(n)=\binom{n}{2} ways to do so. The same is true whenever two of the three shepherds are present. Clearly, whenever only one shepherd works, there are (n1)\binom{n}{1} ways to choose his flock size, and when all shepherds take the day off, there is only one possibility. Therefore, the extended strict order polynomial EP(n,z)\text{E}_{P}^{\circ}(n,z) has the form

EP(n,z)\displaystyle\text{E}_{P}^{\circ}(n,z) =\displaystyle= ΩP(n)z3+(Ω{Fiadh,Aidan}(n)+Ω{Fiadh,Lorcan}(n)+Ω{Aidan,Lorcan}(n))z2\displaystyle\Omega_{P}^{\circ}(n)z^{3}+\left(\Omega_{\{\text{Fiadh},\text{Aidan}\}}^{\circ}(n)+\Omega_{\{\text{Fiadh},\text{Lorcan}\}}^{\circ}(n)+\Omega_{\{\text{Aidan},\text{Lorcan}\}}^{\circ}(n)\right)z^{2}
+(Ω{Fiadh}(n)+Ω{Aidan}(n)+Ω{Lorcan}(n))z1+Ω{Fiadh}(n)z0\displaystyle+\left(\Omega_{\{\text{Fiadh}\}}^{\circ}(n)+\Omega_{\{\text{Aidan}\}}^{\circ}(n)+\Omega_{\{\text{Lorcan}\}}^{\circ}(n)\right)z^{1}+\Omega_{\{\text{Fiadh}\}}^{\circ}(n)z^{0}
=\displaystyle= (n3)z3+3(n2)z2+3(n1)z1+(n0)z0\displaystyle\binom{n}{3}z^{3}+3\binom{n}{2}z^{2}+3\binom{n}{1}z^{1}+\binom{n}{0}z^{0}
=\displaystyle= k=03(3k)(nk)zk.\displaystyle\sum_{k=0}^{3}\binom{3}{k}\binom{n}{k}z^{k}.

The very compact expression for EP(n,z)\text{E}_{P}^{\circ}(n,z) given in the last line of Eq. (2) can be obtained directly by applying the following theorem.

Theorem 3.

The extended strict order polynomial is given by

EP(n,z)=w(P)k=0p(delP(w)pk)(n+des(w)k)zk,\text{\emph{E}}_{P}^{\circ}(n,z)=\sum_{w\in\mathcal{L}(P)}\sum_{k=0}^{p}\binom{\text{del}_{P}(w)}{p-k}\binom{n+\text{des}(w)}{k}z^{k}, (4)

where delP(w)\text{del}_{P}(w) denotes the number of deletable labels in ww.

Intuitively speaking, deletable labels can be understood to be the entries of ww which are not essential for distinguishing ww from other elements of (P)\mathcal{L}(P). Theorem 3 is based on the fact that every element of QP(Q)\bigcup_{Q\subset P}\mathcal{L}(Q) can be uniquely associated with some element of (P)\mathcal{L}(P). Formally speaking, it is possible to define an equivalence relation \sim on QP(Q)\bigcup_{Q\subset P}\mathcal{L}(Q) such that w(P)[w]=QP(Q)\bigcup_{w\in\mathcal{L}(P)}\left[w\right]_{\sim}=\bigcup_{Q\subset P}\mathcal{L}(Q). This concept is illustrated below in Examples 4 and 5. The proof of Theorem 3 will be given at the end of this paper after formalizing the concept of deletable labels and proving some technical lemmata.

Example 4.

Let us consider the lattice P=𝟐×𝟐P=\boldsymbol{2}\times\boldsymbol{2}, for which (P)={1234,1324}\mathcal{L}(P)=\left\{1234,1324\right\} and QP(Q)={,1,2,3,4,12,13,14,23,24,34,32,123,124,134,234,324,132,1234,1324}\bigcup_{Q\subset P}\mathcal{L}(Q)=\left\{\varnothing,1,2,3,4,12,13,14,23,24,34,32,123,124,134,234,324,132,1234,1324\right\}. Our results allow us to partition the set QP(Q)\bigcup_{Q\subset P}\mathcal{L}(Q) of linear extensions into two equivalence classes [1234]\left[1234\right]_{\sim} and [1324]\left[1324\right]_{\sim}:

[Uncaptioned image]

The first family, originating from the linear extension 12341234, is characterized by zero descents (des(w)=0\text{des}(w)=0) and zero non-deletable elements (delP(w)=40\text{del}_{P}(w)=4-0), and thus contains (40k0)\binom{4-0}{k-0} sequences of each length kk. The second family, originating from the linear extension 13241324, is characterized by one descent (des(w)=1\text{des}(w)=1) and two non-deletable elements 3 and 2 (delP(w)=42\text{del}_{P}(w)=4-2), and thus contains (42k2)\binom{4-2}{k-2} sequences of each length kk. Consequently, the extended strict order polynomial is given by

EP(n,z)=k=04((40k0)(nk)+(42k2)(n+1k))zk.\text{E}_{P}^{\circ}(n,z)=\sum_{k=0}^{4}\left(\binom{4-0}{k-0}\binom{n}{k}+\binom{4-2}{k-2}\binom{n+1}{k}\right)z^{k}.
Example 5.

Let us consider the poset P={a,b,c}P=\left\{a,b,c\right\} of three non-comparable elements. We have (P)={123,132,213,231,312,321}\mathcal{L}(P)=\left\{123,132,213,231,312,321\right\} and QP(Q)={,1,2,3,12,21,13,31,23,32,123,132,213,231,312,321}\bigcup_{Q\subset P}\mathcal{L}(Q)=\left\{\varnothing,1,2,3,12,21,13,31,23,32,123,132,213,231,312,321\right\}. Our results allow us to classify the linear extensions in QP(Q)\bigcup_{Q\subset P}\mathcal{L}(Q) into six families

[Uncaptioned image]

each of which is characterized by a pair of numbers (des(w),delP(w))\left(\text{des}(w),\text{del}_{P}(w)\right) specified above. The extended strict order polynomial is given by

EP(n,z)=k=03((3k)(nk)+3(32k2)(n+1k)+(33k3)(n+1k)+(33k3)(n+2k))zk.\text{E}_{P}^{\circ}(n,z)=\sum_{k=0}^{3}\left(\binom{3}{k}\binom{n}{k}+{\color[rgb]{0.80078125,0,0.80078125}3}\binom{3-2}{k-2}\binom{n+1}{k}+\binom{3-3}{k-3}\binom{n+1}{k}+\binom{3-3}{k-3}\binom{n+2}{k}\right)z^{k}.

3 Classification of linear extensions

Definition 6.

Consider a sequence w=w1w2wpw=w_{1}w_{2}\ldots w_{p} which represents a linear extension σ\sigma of PP. A label wiw_{i} is deletable if

  • 1)1)

    neither i1i-1 nor ii is a descent, and

  • 2)2)

    at least one the following is true:

    • a)a)

      wj<wiw_{j}<w_{i} for any j(0,i)j\in(0,i), or

    • b)b)

      there exists a label wkw_{k} with ω1(wk)<Pω1(wi)\omega^{-1}(w_{k})<_{P}\omega^{-1}(w_{i}) such that wj<wiw_{j}<w_{i} for all j(k,i)j\in(k,i).

The set of deletable elements of ww is denoted by DelP(w)\text{Del}_{P}(w), and its cardinality by delP(w)=#DelP(w)\text{del}_{P}(w)=\#\text{Del}_{P}(w). Labels that are not deletable from ww are called fixed in ww.

Loosely speaking, a label is fixed if it contributes to a descent, or if it appears after a descent even though it would also be allowed to appear in front of it. An inclined reader might have already noticed that every deletable element appears in a uniquely defined position of ww, which may be described as ,,as early as possible without interfering with the descent pattern”. In other words, removing a deletable element wiw_{i} from ww and reinserting it at any earlier position cannot result in a linear extension with the same labels involved in the descent pairs. This concept constitutes the main idea behind associating a linear extension vQP(Q)v\in\bigcup_{Q\subset P}\mathcal{L}(Q) with a unique linear extension w(P)w\in\mathcal{L}(P) developed in detail later in this communication.

Let us now demonstrate how Definition 6 can be used in a direct manner to identify deletable and fixed labels in linear extensions. A useful, easy-to-use-in-practice graphical reinterpretation of Definition 6 is introduced in Example 9.

Example 7.

Consider again the poset P=𝟐×𝟐P=\boldsymbol{2}\times\boldsymbol{2} shown in Fig. 2 and its two linear extensions w=1234w=1234 and w=1324w^{\prime}=1324. Let us first determine which of the labels are deletable from 12341234. We have no descents in 12341234, so the condition 1)1) of Definition 6 is satisfied for all of the labels. We only need to verify condition 2)2) of Definition 6. The label w1=1w_{1}=1 is deletable because the interval (0,1)\left(0,1\right) is empty, and therefore condition 2a)2a) is vacuously satisfied. Since w1<w2<w3<w4w_{1}<w_{2}<w_{3}<w_{4}, we find also for the remaining labels wi=2,3,4w_{i}=2,3,4 that condition 2a)2a) is satisfied. This shows that all four labels in the linear extension 12341234 are deletable: DelP(1234)={1,2,3,4}\text{Del}_{P}(1234)=\left\{1,2,3,4\right\} and delP(1234)=4\text{del}_{P}(1234)=4. This is not the case for the linear extension 13241324. The labels 33 and 22 are fixed by condition 1)1) because the position i=2i=2 is a descent. The labels 11 and 44 can be shown to be deletable in exactly the same way as for 12341234. Consequently, in the linear extension 13241324 the set of deletable elements is DelP(1324)={1,4}\text{Del}_{P}(1324)=\left\{1,4\right\} and delP(1324)=2\text{del}_{P}(1324)=2.

Refer to caption
Figure 2: Hasse diagram of two posets P=𝟐×𝟐P=\boldsymbol{2}\times\boldsymbol{2} and P=𝟑×𝟑P=\boldsymbol{3}\times\boldsymbol{3} together with (natural) labelings ω\omega.
Example 8.

Consider the linear extension w=124753689w=124753689 of the poset P=𝟑×𝟑P=\boldsymbol{3}\times\boldsymbol{3} shown in Fig. 2. By condition 1)1) of Definition 6, the labels 3,53,5 and 77 are not deletable; the remaining labels might be deletable if they satisfy condition 2a)2a) or 2b)2b) of Definition 6. Since w1<w2<w3w_{1}<w_{2}<w_{3}, the labels w1=1,w2=2w_{1}=1,w_{2}=2 and w3=4w_{3}=4 are deletable by condition 2a)2a); likewise, due to wj<w8<w9w_{j}<w_{8}<w_{9} for all j=1,,7j=1,\ldots,7, the labels 88 and 99 are deletable by condition 2a)2a). Thus the only remaining label for which the deletability (or non-deletability) is not immediately obvious—and hence the machinery of Definition 6 must be fully put to work—is w7=6w_{7}=6. There exists the label w5=5w_{5}=5 with ω1(5)<Pω1(6)\omega^{-1}(5)<_{P}\omega^{-1}(6) , and for the only label wjw_{j} with j(5,7)={6}j\in(5,7)=\{6\} we find w6=3<w7=6w_{6}=3<w_{7}=6, so by condition 2b)2b), the label w7=6w_{7}=6 is deletable. Therefore, DelP(w)={1,2,4,6,8,9}\text{Del}_{P}(w)=\left\{1,2,4,6,8,9\right\} and delP(w)=6\text{del}_{P}(w)=6.

Example 9.

Linear extensions and their fixed and deletable elements can be visualized and analyzed graphically in the following way. For a given linear extension σ\sigma, plot the Hasse diagram of PP in such a way that each element tit_{i} of PP is represented as a point in a Cartesian plane with the coordinates (σ(ti),ω(ti))=(i,wi)\left(\sigma(t_{i}),\omega(t_{i})\right)=\left(i,w_{i}\right), see Fig. 3(a)(a). The total order implied by the linear extension σ\sigma is easily determined by connecting the elements in the resulting diagram from left to right, as shown using red arrows in Fig. 3(b)(b). The fixed and deletable labels can now be identified in a graphical way, as illustrated in Fig. 3(c)(c): Whenever ii is a descent, i.e. whenever an element tit_{i} (in the position (i,wi)\left(i,w_{i}\right)) is displayed above ti+1t_{i+1} (in the position (i+1,wi+1)\left(i+1,w_{i+1}\right)), both wiw_{i} and wi+1w_{i+1} are fixed (compare condition 1)1) of Definition 6); this is marked by coloring the corresponding elements. The line connecting the points (i,wi)\left(i,w_{i}\right) and (i+1,wi+1)\left(i+1,w_{i+1}\right) then casts a ,,shadow” to the right; and all elements in the shadow have fixed labels as long as they are not covered by any elements in the same shadow including tit_{i} and ti+1t_{i+1} (if they are covered, their labels are deletable by condition 2b)2b)).

Refer to caption

(a)\,\,\,\,\,\,\,\,\,\,\,\,\,(a)

Refer to caption

(b)\,\,\,\,\,\,\,\,\,\,\,\,\,(b)

Refer to caption

(c)\,\,\,\,\,\,\,\,\,\,\,\,\,(c)

Figure 3: Graphical representation of the linear extension σ\sigma represented by w=124735869w=124735869. (a)(a) Hasse diagram plotted in a Cartesian plane with each element tit_{i} of PP displayed at the coordinates (σ(ti),ω(ti))=(i,wi)\left(\sigma(t_{i}),\omega(t_{i})\right)=\left(i,w_{i}\right). (b)(b) Red arrows depict the total order implied by the linear extension σ\sigma. (c)(c) Graphical identification of elements with fixed labels (displayed as colored circles): Every descent ii fixes the labels wiw_{i} and wi+1w_{i+1}. Further, the line between the two elements tit_{i} and ti+1t_{i+1} involved in a descent casts a shadow to the right, and the labels of any elements caught in the shadow—and not covered by other elements inside the same shadow, including tit_{i} and ti+1t_{i+1}—are also fixed. Elements with deletable labels remain displayed as white circles.
Refer to caption

(a)\,\,\,\,\,\,\,\,\,\,\,\,\,(a)

Refer to caption

(b)\,\,\,\,\,\,\,\,\,\,\,\,\,(b)

Refer to caption

(c)\,\,\,\,\,\,\,\,\,\,\,\,\,(c)

Refer to caption

(d)\,\,\,\,\,\,\,\,\,\,\,\,\,(d)

Refer to caption

(e)\,\,\,\,\,\,\,\,\,\,\,\,\,(e)

Refer to caption

(f)\,\,\,\,\,\,\,\,\,\,\,\,\,(f)

Figure 4: Graphical representation of six linear extensions of the poset P=𝟑×𝟑P=\boldsymbol{3}\times\boldsymbol{3} shown in Fig. 2. Elements with fixed labels are shown as colored circles.
Example 10.

Fig. 4 shows six of the linear extensions of the poset P=𝟑×𝟑P=\boldsymbol{3}\times\boldsymbol{3} shown in Fig. 2. The introduced graphical representation of each of the linear extensions allows us to identify easily the sets of deletable labels: (a)(a) DelP(w)=[ 9]\text{Del}_{P}(w)=[\,9\,], (b)(b) DelP(w)=[ 9]{2,4}\text{Del}_{P}(w)=[\,9\,]\setminus\left\{2,4\right\}, (c)(c) DelP(w)=[ 9]{3,7}\text{Del}_{P}(w)=[\,9\,]\setminus\left\{3,7\right\}, (d)(d) DelP(w)=[ 9]{3,5,7}\text{Del}_{P}(w)=[\,9\,]\setminus\left\{3,5,7\right\}, (e)(e) DelP(w)=[ 9]{3,5,7}\text{Del}_{P}(w)=[\,9\,]\setminus\left\{3,5,7\right\}, (f)(f) DelP(w)={1,9}\text{Del}_{P}(w)=\left\{1,9\right\}. Note that in cases (b)(e)(b)-(e), there are elements in the shadow which are not colored because they are covered by another element in the same shadow – that is, elements that are deletable by condition 2b)2b) of Definition 6.

We are now ready to investigate formally the relation between the linear extensions of PP and the linear extensions of its subposets PDP\setminus D.

3.1 Correspondence between linear extensions of a poset and of its subsets

Deleting deletable elements does not affect the number of descents:

Lemma 11.

Consider a linear extension w(P)w\in\mathcal{L}(P) and a subsequence v=wDv=w\setminus D, where DdelP(w)D\subset\text{\emph{del}}_{P}(w). Then, des(w)=des(v)\text{\emph{des}}(w)=\text{\emph{des}}(v).

Proof.

Let us augment the linear extension ww with two auxiliary fixed labels w0=0w_{0}=0 and wp+1=p+1w_{p+1}=p+1. Then any deletable label of ww is located between two fixed labels wiw_{i} and wjw_{j}, which can be selected in such a way that all the labels wi+1,,wj1w_{i+1},\ldots,w_{j-1} in between are deletable. If there is any k(i,j)k\in(i,j) such that wk>wk+1w_{k}>w_{k+1}, kk would be a descent in ww and wkw_{k} and wk+1w_{k+1} would be fixed according to condition 1)1) of Def. 6, contradicting the choice of wiw_{i} and wjw_{j}. Therefore, we have wi<wi+1<<wj1<wjw_{i}<w_{i+1}<\ldots<w_{j-1}<w_{j}. Every deletable element belongs to such an interval containing monotonously increasing deletable labels flanked by two fixed labels, therefore constructing v=wDv=w\setminus D by deleting any deletable elements from ww does not remove or introduce any descents. ∎

The following two lemmata establish the correspondence between the elements of (PD)\mathcal{L}(P\setminus D) and the elements of (P)\mathcal{L}(P) .

Lemma 12.

Let D[p]D\subset[\,p\,], and let Q=PDQ=P\setminus D be a subposet of PP. For every linear extension vv of QQ there exists exactly one linear extension ww of PP such that v=wDv=w\setminus D and DdelP(w)D\subset\text{\emph{del}}_{P}(w).

Proof.

Denote in the following by q=#Q=p#Dq=\#Q=p-\#D the length of vv. For the sake of brevity of the following expositions, assume during this proof that v0=0v_{0}=0 and vq+1=p+1v_{q+1}=p+1. Let us attempt to construct a sequence ww of elements of [p]\left[\,p\,\right] such that

  • a)a)

    v=wDv=w\setminus D,

  • b)b)

    w(P)w\in\mathcal{L}(P),

  • c)c)

    DDelP(w)D\subset\text{Del}_{P}(w).

A sequence ww can only satisfy condition a)a) if it contains the labels v1,,vqv_{1},\ldots,v_{q} appearing in the same order as in vv, preceded, interleaved, and/or succeeded by the elements of D.D. Let us denote by D0D^{0} the set of elements of DD appearing in ww before v1v_{1}, by D1D^{1} the set of elements appearing between v1v_{1} and v2v_{2}, and so on. Clearly, DD is a disjoint union of the subsets D0,D1,,DqD^{0},D^{1},\ldots,D^{q}. We can thus construct a sequence ww in two steps:

  • Step 1:

    Partition DD into q+1q+1 (possibly empty) subsets D0,D1,,DqD^{0},D^{1},\ldots,D^{q} containing m0,m1,,mqm_{0},m_{1},\ldots,m_{q} elements, respectively.

  • Step 2:

    Arrange the elements of each subset DiD^{i} into a subsequence d1id2idmiid_{1}^{i}d_{2}^{i}\ldots d_{m_{i}}^{i} and form a sequence ww by concatenating the labels in vv and the consecutive subsequences d10d20dm00,,d1qd2qdmqqd_{1}^{0}d_{2}^{0}\ldots d_{m_{0}}^{0},\ldots,d_{1}^{q}d_{2}^{q}\ldots d_{m_{q}}^{q} in the following way

    w=d10d20dm00v1d11d21dm11v2vqd1qd2qdmqq.w=d_{1}^{0}d_{2}^{0}\ldots d_{m_{0}}^{0}v_{1}d_{1}^{1}d_{2}^{1}\ldots d_{m_{1}}^{1}v_{2}\ldots v_{q}d_{1}^{q}d_{2}^{q}\ldots d_{m_{q}}^{q}.

Obviously, many different sequences ww can be constructed in this way by choosing different partitionings of DD and by selecting distinct orders of the elements in each DiD^{i}; we show during the following construction process that the conditions b)b) and c)c) restrict this abundance to a single, unique sequence ww.

Every dDd\in D must be inserted in such a way that that wi1<wid<wi+1w_{i-1}<w_{i}\equiv d<w_{i+1}, otherwise, dd would—by condition 1)1) of Definition 6—not be deletable from ww, thus violating condition c)c). Therefore, the subsequence d1id2idmiid_{1}^{i}d_{2}^{i}\ldots d_{m_{i}}^{i} inserted between viv_{i} and vi+1v_{i+1} must satisfy vi<d1i<d2i<<dmii<vi+1v_{i}<d_{1}^{i}<d_{2}^{i}<\ldots<d_{m_{i}}^{i}<v_{i+1}. This shows that, in Step 2, when we augment vv with the elements of a subset DiD^{i}, the only choice is to arrange these elements into a monotonously increasing sequence before doing so. Moreover, this requirement seriously reduces the number of allowed partitions of DD into subsets DiD^{i}, as each dDid\in D^{i} needs to satisfy the condition vi<d<vi+1v_{i}<d<v_{i+1}.

Consider an element dDd\in D. Let us now narrow down the family of subsets DiD^{i} into which the element dd may be placed. Let jd=max{j[q]|ω1(vj)<Pω1(d)}j_{d}=\text{max}\left\{j\in[\,q\,]\,|\,\omega^{-1}(v_{j})<_{P}\omega^{-1}(d)\right\}, or jd=0j_{d}=0 if this set is empty. In order to not violate condition b)b), dd must be in some DiD^{i} with jdij_{d}\leq i. Denote by Id={i|ijd,vi<d<vi+1}I_{d}=\left\{i\,|\,i\geq j_{d},\,v_{i}<d<v_{i+1}\right\} the set of possible choices for ii limited by the so far derived conditions ijdi\geq j_{d} and vi<d<vi+1v_{i}<d<v_{i+1}. The set IdI_{d} is nonempty: It follows from the order-preserving nature of ω\omega that vjd<d<vq+1v_{j_{d}}<d<v_{q+1}, so there must be at least one value of ii with jdiqj_{d}\leq i\leq q such that vi<d<vi+1v_{i}<d<v_{i+1}. Let id=min Idi_{d}=\text{min }I_{d}.

We will now show by reductio ad absurdum that placing dd into a subset other than DidD^{i_{d}} leads to a violation of condition c)c). (Recollect that every deletable element appears in ww ,,as early as possible”.) Assume that dDid\in D^{i} with iIdi\in I_{d} and i>idi>i_{d}. Then, by definition of IdI_{d}, we know that d<vid+1d<v_{i_{d}+1}. In order for dd to be deletable from the sequence ww, there must be an element e[p]e\in[\,p\,] such that ω1(e)<Pω1(d)\omega^{-1}(e)<_{P}\omega^{-1}(d) and which appears in ww between vid+1v_{i_{d}+1} and dd. Denote by EE the set of such elements: E={e[p]|ω1(e)<Pω1(d),σ(ω1(vid+1))<σ(ω1(e))<σ(ω1(d))}E=\left\{e\in[\,p\,]\,|\,\omega^{-1}(e)<_{P}\omega^{-1}(d),\sigma(\omega^{-1}(v_{i_{d}+1}))<\sigma(\omega^{-1}(e))<\sigma(\omega^{-1}(d))\right\}, where σ\sigma denotes the map σ:P𝒑\sigma:P\rightarrow\boldsymbol{p}, wiiw_{i}\mapsto i implied by ww. If there is an eEe\in E with eDe\notin D, then ee must appear in vv at some position kk, evke\equiv v_{k}. Since ω1(e)<ω1(d)\omega^{-1}(e)<\omega^{-1}(d), according to the definitions of jdj_{d} and idi_{d} we find that kjdidk\leq j_{d}\leq i_{d}, in contradiction with the requirement that vid+1v_{i_{d}+1} precede e=vke=v_{k} in vv. Therefore, eDe\in D and thus EDE\subset D. Consider now the element c=min Ec=\text{min }E. It follows from ω1(c)<Pω1(d)\omega^{-1}(c)<_{P}\omega^{-1}(d) that c<d<vid+1c<d<v_{i_{d}+1}. In order for cc to be deletable from the finished sequence ww, there must be an element e[p]e^{\prime}\in[\,p\,] such that ω1(e)<Pω1(c)\omega^{-1}(e^{\prime})<_{P}\omega^{-1}(c) and which appears in ww between vid+1v_{i_{d}+1} and cc. Because of ω1(c)<ω1(d)\omega^{-1}(c)<\omega^{-1}(d) and since cc must precede dd in ww, the aforementioned element ee^{\prime} must be in EE, and therefore min E=c<e\text{min }E=c<e^{\prime}. At the same time, since ω\omega is a natural labeling, ω1(e)<Pω1(c)\omega^{-1}(e^{\prime})<_{P}\omega^{-1}(c) implies e<ce^{\prime}<c. This contradiction shows that cc cannot be deletable from ww, cDelP(w)c\notin\text{Del}_{P}(w). However we have found before that cDc\in D, which means that the assumption i>idi>i_{d} leads to a violation of condition c)c). Therefore, we must have iidi\leq i_{d}. Since idi_{d} is defined as the minimum allowed value of i,i, we have i=idi=i_{d}.

To summarize, we have shown until now that the only way to construct a sequence ww in a way that does not contradict conditions a)c)a)-c) is to follow the construction introduced above, which can be described in the following way:

  • Step 1:

    For every dDd\in D, let

    jd\displaystyle\phantom{\text{and }}j_{d} =\displaystyle= max({j[q]|ω1(vj)<Pω1(d)}{0})\displaystyle\text{max}\left(\left\{j\in[\,q\,]\,|\,\omega^{-1}(v_{j})<_{P}\omega^{-1}(d)\right\}\cup\left\{0\right\}\right) (5)
    and id\displaystyle\text{and }i_{d} =\displaystyle= min({i|jdiq,vi<d<vi+1}),\displaystyle\text{min}\left(\left\{i\,|\,j_{d}\leq i\leq q,\,v_{i}<d<v_{i+1}\right\}\right), (6)

    and assign dd to the set DidD^{i_{d}}.

  • Step 2:

    For every 0iq0\leq i\leq q, insert between viv_{i} and vi+1v_{i+1} the elements of DiD^{i} in increasing order:

    w=d10d20dm00v1d11d21dm11v2vqd1qd2qdmqq,w=d_{1}^{0}d_{2}^{0}\ldots d_{m_{0}}^{0}v_{1}d_{1}^{1}d_{2}^{1}\ldots d_{m_{1}}^{1}v_{2}\ldots v_{q}d_{1}^{q}d_{2}^{q}\ldots d_{m_{q}}^{q},

    where dqiDid_{q}^{i}\in D^{i} and vi<d1i<d2i<<dmii<vi+1v_{i}<d_{1}^{i}<d_{2}^{i}<\ldots<d_{m_{i}}^{i}<v_{i+1}.

It remains to be demonstrated that the sequence ww uniquely defined in this way indeed satisfies all conditions a)c)a)-c). Condition a)a) is satisfied by construction.

Next, let us verify that ww satisfies condition b)b). Consider two arbitrary elements s,tPs,t\in P such that s<Pts<_{P}t. Each of their labels ω(s)\omega(s) and ω(t)\omega(t) can be in DD or in [p]D[\,p\,]\setminus D. For each case, we have to show that ω(s)\omega(s) precedes ω(t)\omega(t) in ww.

  • If ω(s),ω(t)[p]D\omega(s),\omega(t)\in[\,p\,]\setminus D, then ω(s)vi\omega(s)\equiv v_{i} and ω(t)vj\omega(t)\equiv v_{j} for some i,j[q]i,j\in[\,q\,]. Since QQ is an induced subposet of PP, we have s<Qts<_{Q}t, and therefore we know that i<ji<j, i.e., viω(s)v_{i}\equiv\omega(s) precedes vjω(t)v_{j}\equiv\omega(t) in vv. Then, by construction, ω(s)\omega(s) precedes ω(t)\omega(t) also in ww.

  • If ω(s)[p]D\omega(s)\in[\,p\,]\setminus D and ω(t)D\omega(t)\in D, then ω(s)vk\omega(s)\equiv v_{k} for some k[q]k\in[\,q\,]. Step 1 defines two numbers jω(t)j_{\omega(t)} and iω(t)i_{\omega(t)}. Since s<Pts<_{P}t, kk is in {j[q]|ω1(vj)<Pt}\left\{j\in[q]\,|\,\omega^{-1}(v_{j})<_{P}t\right\}, and thus by Eq. (5) kjω(t)k\leq j_{\omega(t)}. From Eq. (6) it is clear that jω(t)iω(t)j_{\omega(t)}\leq i_{\omega(t)}. Consequently, the label ω(t)\omega(t) is assigned to Diω(t)D^{i_{\omega(t)}} with kiω(t)k\leq i_{\omega(t)}, which means that ω(t)\omega(t) appears in ww after ω(s)vk\omega(s)\equiv v_{k}.

  • If ω(s)D\omega(s)\in D and ω(t)[p]D\omega(t)\in[\,p\,]\setminus D, then ω(t)vk\omega(t)\equiv v_{k} for some k[q]k\in[\,q\,]. Any vl[p]Dv_{l}\in[\,p\,]\setminus D with ω1(vl)<Ps\omega^{-1}(v_{l})<_{P}s also satisfies ω1(vl)<Ps<Pt=ω1(vk)\omega^{-1}(v_{l})<_{P}s<_{P}t=\omega^{-1}(v_{k}), and therefore l<kl<k. Therefore, application of Step 11 to ω(s)\omega(s) results in jω(s)<kj_{\omega(s)}<k and, due to the fact that ω(s)<ω(t)\omega(s)<\omega(t), we have iω(s)<ki_{\omega(s)}<k. Thus, ω(s)\omega(s) is assigned to a Diω(s)D^{i_{\omega(s)}} with iω(s)<ki_{\omega(s)}<k, and therefore it appears in ww before ω(t)\omega(t).

  • If ω(s),ω(t)D\omega(s),\omega(t)\in D, then for any vkv_{k} with ω1(vk)<Ps\omega^{-1}(v_{k})<_{P}s, it follows directly that ω1(vk)<Pt\omega^{-1}(v_{k})<_{P}t. Therefore, in Step 11, we find jω(s)jω(t)j_{\omega(s)}\leq j_{\omega(t)}, and as a result, in addition to jω(s)iω(s)j_{\omega(s)}\leq i_{\omega(s)} (due to Eq. (6)) we also know that jω(s)iω(t)j_{\omega(s)}\leq i_{\omega(t)}. By construction of jω(s)j_{\omega(s)} and iω(t)i_{\omega(t)} as well as the order-preserving nature of ω\omega, we find vjω(s)<ω(s)<ω(t)<viω(t)+1v_{j_{\omega(s)}}<\omega(s)<\omega(t)<v_{i_{\omega(t)}+1}. Therefore, there must be at least one value of ii in the interval jω(s),,iω(t)j_{\omega(s)},\ldots,i_{\omega(t)} such that vi<ω(s)<vi+1v_{i}<\omega(s)<v_{i+1}. It follows that iω(s)iω(t)i_{\omega(s)}\leq i_{\omega(t)}. If iω(s)<iω(t)i_{\omega(s)}<i_{\omega(t)}, then obviously ω(s)\omega(s) appears in ww before viω(t)v_{i_{\omega(t)}}, which in turn appears before ω(t)\omega(t). Finally, even if iω(s)=iω(t)i_{\omega(s)}=i_{\omega(t)}, in Step 22 the elements of each DiD^{i} are inserted into the sequence ww in increasing order, so in any case ω(s)\omega(s) will be inserted before ω(t)\omega(t).

We have shown that the constructed sequence ww satisfies the condition b)b), w(P)w\in\mathcal{L}(P).

Finally let us verify that ww satisfies condition c)c). Consider an element dDd\in D which is inserted into ww at some position kk, thus dwkd\equiv w_{k}. We have ensured during the construction process that wk1<dwk<wk+1w_{k-1}<d\equiv w_{k}<w_{k+1}, therefore condition 1)1) of Def. 6 is satisfied. If wj<wkw_{j}<w_{k} for all j(0,k)j\in(0,k), then condition 2a)2a) of Def. 6 is satisfied and dd is deletable in ww. Otherwise, the set L={i|i<k and wi>wk}L=\left\{i\,|\,i<k\text{ and }w_{i}>w_{k}\right\} is nonempty. Let l=max Ll=\text{max }L. It is clear that wi<wkw_{i}<w_{k} for all i(l,k)i\in(l,k). Thus, if we can find a value of j(l,k)j\in(l,k) such that ω1(wj)<Pω1(wk)\omega^{-1}(w_{j})<_{P}\omega^{-1}(w_{k}), then condition 2b)2b) of Def. 6 is satisfied. We show below that indeed this is the case.

It follows directly from the definition of ll that wl>wkw_{l}>w_{k} and simultaneously wk>wl+1w_{k}>w_{l+1} (since l+1(l,k)l+1\in\left(l,k\right)); therefore wl>wl+1w_{l}>w_{l+1}, and ll is a descent. We already have seen that condition 1)1) of Def. 6 is satisfied for all the elements of DD; therefore wlw_{l} and wl+1w_{l+1} are not in DD, but appear somewhere in the original sequence vv, in the form wlvl~w_{l}\equiv v_{\tilde{l}} and wl+1vl~+1w_{l+1}\equiv v_{\tilde{l}+1} for some l~[q]\tilde{l}\in[\,q\,]. Since l+1<kl+1<k, during Step 1 dd must have been assigned to some DidD^{i_{d}} with l~<id\tilde{l}<i_{d}.

Let us assume that jd<l~j_{d}<\tilde{l}; we will see that this assumption leads to a contradiction. From ω1(vjd)<Pω1(d)\omega^{-1}(v_{j_{d}})<_{P}\omega^{-1}(d) it follows that vjd<dv_{j_{d}}<d. By construction of l~\tilde{l}, we have d<vl~d<v_{\tilde{l}}. Therefore, if jd<l~j_{d}<\tilde{l}, then there must be a i(jd1,l~)i\in(j_{d}-1,\tilde{l}) such that vi<d<vi+1v_{i}<d<v_{i+1}. Then, by definition of idi_{d}, we would have idi<l~i_{d}\leq i<\tilde{l}, in contradiction with l~<id\tilde{l}<i_{d}. Therefore, the assumption jd<l~j_{d}<\tilde{l} made at the beginning of this paragraph must be wrong and we have l~jd\tilde{l}\leq j_{d}. Since vl~>dv_{\tilde{l}}>d, we find ω1(vl~)Pω1(d)\omega^{-1}(v_{\tilde{l}})\nless_{P}\omega^{-1}(d), and therefore (by Eq. (5)) jdl~j_{d}\neq\tilde{l}. This reasoning shows that l~<jd\tilde{l}<j_{d}.

The entry vjdv_{j_{d}} of vv appears in ww in the form vjdwjv_{j_{d}}\equiv w_{j} at some position j[p]j\in[\,p\,]. Since l~<jdid\tilde{l}<j_{d}\leq i_{d} and elements of vv appear in the same order in ww, we find l<j<kl<j<k. As we have shown earlier, by construction of ll, we have wm<wkw_{m}<w_{k} for all m(l,k)m\in(l,k) (and thus especially for all m(j,k)m\in(j,k)); and by construction of jj, we have ω1(wj)<Pω1(wk)\omega^{-1}(w_{j})<_{P}\omega^{-1}(w_{k}). Therefore, condition 2b)2b) of Def. 6 is satisfied. It follows that every dDd\in D is deletable in ww, meaning that DDelP(w)D\subset\text{Del}_{P}(w).

We have demonstrated that there is exactly one sequence ww of elements of [p]\left[p\right] that satisfies conditions a)c)a)-c) given at the beginning of this proof, i.e., there is exactly one linear extension ww such that v=wDv=w\setminus D and DDelP(w)D\subset\text{Del}_{P}(w). ∎

Example 13.

Let us demonstrate the insertion process described in Steps 1&2 during the proof above. Consider the sequence v=17536(PD)v=17536\in\mathcal{L}(P\setminus D) with the poset P=𝟑×𝟑P=\boldsymbol{3}\times\boldsymbol{3} shown in Fig. 2 and the set of deleted elements D={2,4,8,9}D=\left\{2,4,8,9\right\}. For the labels d=2d=2 and 44, we find for the set in Eq. (5) {j[ 5]|ω1(vj)<Pω1(d)}={j[ 5]|vj{1}}={1}\left\{j\in[\,5\,]\,|\,\omega^{-1}(v_{j})<_{P}\omega^{-1}(d)\right\}=\left\{j\in[\,5\,]\,|\,v_{j}\in\left\{1\right\}\right\}=\left\{1\right\}, and thus j2=j4=1j_{2}=j_{4}=1. Since v1=1<2,4<v2=7v_{1}=1<2,4<v_{2}=7, in Eq. (6) we find i2=i4=1i_{2}=i_{4}=1. Therefore, the labels 22 and 44 are assigned to the subset D1D^{1}, and will be inserted into the sequence vv between v1=1v_{1}=1 and v2=7v_{2}=7. For the label 88, we find that in Eq. (5), {j[ 5]|ω1(vj)<Pω1(8)}={j[ 5]|vj{5,7}}={2,3}\left\{j\in[\,5\,]\,|\,\omega^{-1}(v_{j})<_{P}\omega^{-1}(8)\right\}=\left\{j\in[\,5\,]\,|\,v_{j}\in\left\{5,7\right\}\right\}=\left\{2,3\right\}, and thus j8=3j_{8}=3. Since however the label 88 is larger than any of the following labels in vv, v3=5v_{3}=5, v4=3v_{4}=3 and v5=6v_{5}=6, we find in Eq. (6) that {i|j8=3i5,vi<8<vi+1}={5}\left\{i\,|\,j_{8}=3\leq i\leq 5,\,v_{i}<8<v_{i+1}\right\}=\left\{5\right\} and therefore i8=5i_{8}=5. Finally, for the label 99, we find that in Eq. (5), {j[ 5]|ω1(vj)<Pω1(9)}={j[ 5]|vj{6,8}}={5}\left\{j\in[\,5\,]\,|\,\omega^{-1}(v_{j})<_{P}\omega^{-1}(9)\right\}=\left\{j\in[\,5\,]\,|\,v_{j}\in\left\{6,8\right\}\right\}=\left\{5\right\}, and thus j9=5j_{9}=5 and i9=5i_{9}=5. To summarize, in Step 1, we split the set D={2,4,8,9}D=\left\{2,4,8,9\right\} into the subsets D0=D^{0}=\varnothing, D1={2,4}D^{1}=\left\{2,4\right\}, D2=D^{2}=\varnothing, D3=D^{3}=\varnothing, D4=D^{4}=\varnothing and D5={8,9}D^{5}=\left\{8,9\right\}. In Step 2, the elements of each subset are arranged into a growing sequence, specifically d11d21=24d_{1}^{1}d_{2}^{1}=24 and d15d25=89d_{1}^{5}d_{2}^{5}=89, and inserted into vv:

[Uncaptioned image]

The resulting sequence w=124753689w=124753689 was previously considered and illustrated in Fig. 4 (e)(e).

Lemma 14.

Let ww be a linear extension of PP with the set of deletable elements DelP(w)\text{Del}_{P}(w). For every set DDelP(w)D\subset\text{Del}_{P}(w), the sequence wDw\setminus D is a linear extension of PDP\setminus D.

Proof.

Consider two elements s,tPDs,t\in P\setminus D with s<PDts<_{P\setminus D}t. In order to show that wDw\setminus D is a linear extension of PDP\setminus D, we have to demonstrate that ω(s)\omega(s) appears in vv before ω(t)\omega(t). Since PDP\setminus D is an induced subposet of PP, it follows from s<PDts<_{P\setminus D}t that s<Pts<_{P}t, and since ww is a linear extension, this implies that ω(s)\omega(s) precedes ω(t)\omega(t) in ww. Clearly then, by construction of vv, ω(s)\omega(s) also precedes ω(t)\omega(t) in vv. ∎

By combining the previous two lemmata, we find that

Lemma 15.

The union of the Jordan-Hölder sets of all subposets of PP is given by

QP(Q)=w(P)DDelP(w){wD}.\bigcup_{Q\subset P}\mathcal{L}(Q)=\bigcup_{w\in\mathcal{L}(P)}\bigcup_{D\subset\text{Del}_{P}(w)}\left\{w\setminus D\right\}.
Proof.

Consider first a set D[p]D\subset[\,p\,] and the corresponding subposet of PP given by PDP\setminus D. It follows directly from Lemmata 12 and 14 that the collection of linear extensions of PDP\setminus D is can be written as

(PD)=w(P)for whichDDelP(w){wD}.\mathcal{L}(P\setminus D)=\bigcup_{\begin{subarray}{c}w\in\mathcal{L}(P)\\ \text{for which}\\ D\subset\text{Del}_{P}(w)\end{subarray}}\left\{w\setminus D\right\}.

Therefore, the set of linear extensions of subposets of PP is given by

QP(Q)=D[p](PD)=D[p]w(P)for whichDDelP(w){wD}=w(P)DDelP(w){wD}.\bigcup_{Q\subset P}\mathcal{L}(Q)=\bigcup_{D\subset[\,p\,]}\mathcal{L}(P\setminus D)=\bigcup_{D\subset[\,p\,]}\bigcup_{\begin{subarray}{c}w\in\mathcal{L}(P)\\ \text{for which}\\ D\subset\text{Del}_{P}(w)\end{subarray}}\left\{w\setminus D\right\}=\bigcup_{w\in\mathcal{L}(P)}\bigcup_{D\subset\text{Del}_{P}(w)}\left\{w\setminus D\right\}.

4 Proof of Theorem 3

We are now ready to combine the so far derived lemmata into the derivation of the closed from of the extended order polynomial given in the first section.

Proof.

(of Theorem 3) By application of the Definitions given in Eqs. (1) and (2) as well as (in line 3) Lemmata 11 and 15, we find

EP(n,z)\displaystyle\text{E}_{P}^{\circ}(n,z) =\displaystyle= QPΩQ(n)z#Q\displaystyle\sum_{Q\subset P}\Omega_{Q}^{\circ}(n)z^{\#Q}
=\displaystyle= QPv(Q)(n+des(v)#Q)z#Q\displaystyle\sum_{Q\subset P}\sum_{v\in\mathcal{L}(Q)}\binom{n+\text{des}(v)}{\#Q}z^{\#Q}
=\displaystyle= w(P)DDelP(w)(n+des(w)p#D)zp#D\displaystyle\sum_{w\in\mathcal{L}(P)}\sum_{D\subset\text{Del}_{P}(w)}\binom{n+\text{des}(w)}{p-\#D}z^{p-\#D}
=\displaystyle= w(P)k=0p#{DDelP(w)|#D=k}(n+des(w)pk)zpk\displaystyle\sum_{w\in\mathcal{L}(P)}\sum_{k=0}^{p}\#\left\{D\subset\text{Del}_{P}(w)\,|\,\#D=k\right\}\cdot\binom{n+\text{des}(w)}{p-k}z^{p-k}
=\displaystyle= w(P)k=0p(delP(w)k)(n+des(w)pk)zpk\displaystyle\sum_{w\in\mathcal{L}(P)}\sum_{k=0}^{p}\binom{\text{del}_{P}(w)}{k}\cdot\binom{n+\text{des}(w)}{p-k}z^{p-k}

By inverting the order of summation in the inner sum, we obtain Eq. (4). ∎

5 Perspectives

Our main motivation to develop the extended strict order polynomial EP(n,z)\text{E}_{P}^{\circ}(n,z) introduced in the current communication is its close relation to the Zhang-Zhang polynomial [4, 5, 6] enumerating Clar covers of benzenoid hydrocarbons [7], a topic to which we devoted feverish activity in our laboratory for almost a decade now [8, 9, 10, 11, 12, 13, 14]. Our recent contribution, introducing the interface theory of benzenoids [13, 14], demonstrated that enumeration of Clar covers of a benzenoid 𝑩\boldsymbol{B} can be efficiently achieved by studying distributions of covered interface edges in interfaces of 𝑩\boldsymbol{B}. The relative positions of the covered edges can be expressed in a form of a poset. Without giving too many unnecessary details, we can say that for a regular benzenoid strip 𝑩\boldsymbol{B} of length nn, there exists a poset PP such that the extended strict order polynomial EP(n,z)\text{E}_{P}^{\circ}(n,z) coincides with the Zhang-Zhang polynomial ZZ(𝑩,x)\textrm{ZZ}\left(\boldsymbol{B},x\right) of 𝑩\boldsymbol{B} (with z=x+1z=x+1). A detailed proof of this fact is quite technical and will be announced soon. The announced equivalence between the extended strict order polynomials EP(n,z)\text{E}_{P}^{\circ}(n,z) developed in the current study and the Zhang-Zhang polynomials ZZ(𝑩,x)\textrm{ZZ}\left(\boldsymbol{B},x\right) of regular benzenoid strips 𝑩\boldsymbol{B} allows us to recognize (currently without a formal proof) a large collection of facts about EP(n,z)\text{E}_{P}^{\circ}(n,z) due to the previously discovered facts about the ZZ polynomials. Among others, the following facts are easy to deduce:

  1. 1.

    The chain P=𝒑P=\boldsymbol{p} corresponds to a parallelogram M(p,n)M\left(p,n\right)

    [Uncaptioned image]

    for which the ZZ polynomial is given in form of a hypergeometric function, ZZ(M(m,n),x)=F12[m,n1;x+1]\textrm{ZZ}\left(M\left(m,n\right),x\right)={}_{2}F_{1}\negthinspace\left[\begin{array}[]{c}-m,-n\\ 1\end{array};x+1\right] [15, 8, 16]; consequently, we have

    E𝒑(n,z)=F12[p,n1;z].\text{E}_{\boldsymbol{p}}^{\circ}(n,z)={}_{2}F_{1}\negthinspace\left[\begin{array}[]{c}-p,-n\\ 1\end{array};z\right]. (7)

    This result is also directly obvious from Theorem 3: The Jordan-Hölder set of 𝒑\boldsymbol{p} consists of only one element, (𝒑)={123p}\mathcal{L}(\boldsymbol{p})=\{123\ldots p\}, for which del𝒑(123p)=p\text{del}_{\boldsymbol{p}}(123\ldots p)=p and des𝒑(123p)=0\text{des}_{\boldsymbol{p}}(123\ldots p)=0. Thus, Eq. (4) immediately assumes the form of Eq. (7).

  2. 2.

    The poset PP containing pp non-comparable elements corresponds, according to the interface theory of benzenoids, to a prolate rectangle Pr(p,n)Pr\left(p,n\right)

    [Uncaptioned image]

    for which the ZZ polynomial is given by ZZ(Pr(m,n),x)=(1+n(x+1))m\textrm{ZZ}\left(Pr\left(m,n\right),x\right)=\left(1+n\left(x+1\right)\right)^{m} [6, 17, 18]; consequently, we have

    E[p](n,z)=(1+nz)p.\text{E}_{[\,p\,]}^{\circ}(n,z)=\left(1+nz\right)^{p}. (8)
  3. 3.

    The poset P=𝟐×𝒎P=\boldsymbol{2}\times\boldsymbol{m} corresponds to a hexagonal graphene flake O(2,m,n)O\left(2,m,n\right)

    [Uncaptioned image]

    It follows from the ZZ polynomial ZZ(O(2,m,n),x)\text{ZZ}\left(O\left(2,m,n\right),x\right) [19, 20, 21] that the strict order polynomial has the form of a 2×22\times 2 determinant

    E𝟐×𝒎(n,z)=|F12[m,n1;z]z(m+12)F12[1m,1n3;z]z(n+12)F12[1m,1n3;z]F12[m,n1;z]|\text{E}_{\boldsymbol{2}\times\boldsymbol{m}}^{\circ}(n,z)=\left|\begin{array}[]{ll}\phantom{z\binom{n+1}{2}\,}{}_{2}F_{1}\negthinspace\left[\begin{array}[]{c}\phantom{1}-m,\phantom{1}-n\\ 1\end{array};z\right]&\,\,\,z\binom{m+1}{2}\,{}_{2}F_{1}\negthinspace\left[\begin{array}[]{c}1-m,1-n\\ 3\end{array};z\right]\\ \\ z\binom{n+1}{2}\,{}_{2}F_{1}\negthinspace\left[\begin{array}[]{c}1-m,1-n\\ 3\end{array};z\right]&\,\,\,\phantom{z\binom{m+1}{2}\,}{}_{2}F_{1}\negthinspace\left[\begin{array}[]{c}\phantom{1}-m,\phantom{1}-n\\ 1\end{array};z\right]\end{array}\right| (9)
  4. 4.

    The strict order polynomial for the lattice P=𝒍×𝒎P=\boldsymbol{l}\times\boldsymbol{m} is unknown, following the fact that this poset corresponds to the hexagonal flake O(l,m,n)O\left(l,m,n\right)

    [Uncaptioned image]

    The ZZ polynomial ZZ(O(l,m,n),x)\text{ZZ}\left(O\left(l,m,n\right),x\right) of this structure constitutes the hardest unsolved problem in the theory of ZZ polynomials [17, 10, 19, 22].

  5. 5.

    The fence P=Q(1,m)P=Q(1,m) with mm elements corresponds to a zigzag chain Z(m,n)Z(m,n)

    [Uncaptioned image]

    The expression for ZZ(Z(m,n),x)\text{ZZ}\left(Z\left(m,n\right),x\right)—and consequently, EP(n,z)\text{E}_{P}^{\circ}(n,z)—is given by a very lengthy formula [17, 10, 23], but the associated generating function has the form of a continued fraction [23]

    [Uncaptioned image] (10)

    An analogous generating function with respect to nn is unknown.

The extended order polynomial EP(n,z)\text{E}_{P}^{\circ}(n,z) can be also computed in an efficient fashion directly from Eq. (4) through an algorithm based on a graph of ,,compatible” antichains of PP. Propagating weights through this graph in a certain way yields the extended order polynomial without ever having to construct the entire set (P)\mathcal{L}(P). This algorithm has been implemented in Maple 16 [24] and will be reported later. For the example of the poset P=𝟑×𝟑P=\boldsymbol{3}\times\boldsymbol{3} depicted in Fig. 2, we obtain in this way

E𝟑×𝟑(n,z)\displaystyle\text{E}_{\boldsymbol{3}\times\boldsymbol{3}}^{\circ}(n,z) =\displaystyle= k=09((9k)(nk)+(9(92k2)+(93k3))(n+1k)\displaystyle\sum_{k=0}^{9}\left(\binom{9}{k}\binom{n}{k}+\left(9\binom{9-2}{k-2}+\binom{9-3}{k-3}\right)\binom{n+1}{k}\right.
+((93k3)+17(94k4)+2(95k5))(n+2k)\displaystyle\hphantom{\sum_{k=0}^{9}\,}+\left(\binom{9-3}{k-3}+17\binom{9-4}{k-4}+2\binom{9-5}{k-5}\right)\binom{n+2}{k}
+(2(95k5)+7(96k6)+(97k7))(n+3k)+(97k7)(n+4k))zk.\displaystyle\hphantom{\sum_{k=0}^{9}}\left.\,+\left(2\binom{9-5}{k-5}+7\binom{9-6}{k-6}+\binom{9-7}{k-7}\right)\binom{n+3}{k}+\binom{9-7}{k-7}\binom{n+4}{k}\right)z^{k}.

We suspect that the coefficients el,j(P)e_{l,j}\left(P\right) appearing in EP(n,z)\text{E}_{P}^{\circ}(n,z) in front of the terms (92ljk2lj)(n+lk)\binom{9-2l-j}{k-2l-j}\binom{n+l}{k} are #P-complete to compute, in close analogy to the coefficients e(P)e\left(P\right) corresponding to the number of linear extensions of PP. These coefficients are growing very fast with the size of the poset PP. The largest of the coefficients el,j(𝟑×𝟑)e_{l,j}\left(\boldsymbol{3}\times\boldsymbol{3}\right) is only 17 (as can be easily seen from Eq. (5)), but larger PP are characterized by much greater coefficients, e.g., max(el,j(𝟒×𝟒))=3765\max\text{$\left(e_{l,j}\left(\boldsymbol{4}\times\boldsymbol{4}\right)\right)$=3765}, max(el,j(𝟒×𝟓))=200440\max\text{$\left(e_{l,j}\left(\boldsymbol{4}\times\boldsymbol{5}\right)\right)$=200440}, max(el,j(𝟓×𝟓))=61885401\max\text{$\left(e_{l,j}\left(\boldsymbol{5}\times\boldsymbol{5}\right)\right)$=61885401}, and max(el,j(𝟓×𝟔))=27950114975\max\text{$\left(e_{l,j}\left(\boldsymbol{5}\times\boldsymbol{6}\right)\right)$=}27950114975.

It seems that the introduced here extended strict order polynomial EP(n,z)\text{E}_{P}^{\circ}(n,z) can be immediately generalized to the non-strict case using the reciprocity theorem of Stanley (Corollary 3.15.12 of [1]), but this problem is not pursued here further.

References

  • [1] R. Stanley, Enumerative Combinatorics, volume 1 of The Wadsworth & Brooks/Cole Mathematics Series, Springer US, 2nd edition, 2012.
  • [2] R. P. Stanley, A chromatic-like polynomial for ordered sets, Proc. Second Chapel Hill Conference on Combinatorial Mathematics and Its Applications (1970) 421–427.
  • [3] R. P. Stanley, Ordered structures and partitions, volume 0 of Memoirs of the American Mathematical Society, American Mathematical Society, 1972, ISSN: 0065-9266, 1947-6221 Issue: 119.
  • [4] H. P. Zhang, F. J. Zhang, The Clar covering polynomial of hexagonal systems I, Discrete Appl. Math. 69 (1996) 147–167.
  • [5] F. J. Zhang, H. P. Zhang, Y. T. Liu, The Clar covering polynomial of hexagonal systems II, Chin. J. Chem. 14 (1996) 321–325.
  • [6] H. P. Zhang, F. J. Zhang, The Clar covering polynomial of hexagonal systems III, Discrete Math. 212 (2000) 261–269.
  • [7] E. Clar, The Aromatic Sextet, Wiley, London, 1972.
  • [8] C. P. Chou, H. A. Witek, An algorithm and FORTRAN program for automatic computation of the Zhang–Zhang polynomial of benzenoids, MATCH Commun. Math. Comput. Chem. 68(1) (2012) 3–30.
  • [9] C. P. Chou, H. A. Witek, ZZDecomposer: A graphical toolkit for analyzing the Zhang–Zhang polynomials of benzenoid structures, MATCH Commun. Math. Comput. Chem. 71(3) (2014) 741–764.
  • [10] C. P. Chou, H. A. Witek, Determination of Zhang–Zhang polynomials for various classes of benzenoid systems: Non–heuristic approach, MATCH Commun. Math. Comput. Chem. 72(1) (2014) 75–104.
  • [11] H. A. Witek, G. Moś, C. P. Chou, Zhang–Zhang polynomials of regular 3– and 4–tier benzenoid strips, MATCH Commun. Math. Comput. Chem. 73(2) (2015) 427–442.
  • [12] H. A. Witek, J. Langner, G. Moś, C. P. Chou, Zhang–Zhang polynomials of regular 5–tier benzenoid strips, MATCH Commun. Math. Comput. Chem. 78(2) (2017) 487–504.
  • [13] J. Langner, H. A. Witek, Interface theory of benzenoids, MATCH Commun. Math. Comput. Chem. (Submitted) 84 (2020) 143–176.
  • [14] J. Langner, H. A. Witek, Interface theory of benzenoids: Basic applications, MATCH Commun. Math. Comput. Chem. 84 (2020) 177–215.
  • [15] I. Gutman, B. Borovićanin, Zhang–Zhang polynomial of multiple linear hexagonal chains, Z. Naturforsch. A 61 (2006) 73–77.
  • [16] C. P. Chou, H. A. Witek, Closed–form formulas for the Zhang–Zhang polynomials of benzenoid structures: Chevrons and generalized chevrons, MATCH Commun. Math. Comput. Chem. 72(1) (2014) 105–124.
  • [17] C. P. Chou, Y. T. Li, H. A. Witek, Zhang–Zhang polynomials of various classes of benzenoid systems, MATCH Commun. Math. Comput. Chem. 68(1) (2012) 31–64.
  • [18] C. P. Chou, J.-S. Kang, H. A. Witek, Closed–form formulas for the Zhang–Zhang polynomials of benzenoid structures: Prolate rectangles and their generalizations, Discrete Appl. Math. 198 (2016) 101–108.
  • [19] B. H. He, J. Langner, H. A. Witek, Hexagonal flakes as fused parallelograms: A determinantal formula for Zhang-Zhang polynomials of the benzenoids, J. Chin. Chem. Soc. (submitted) (2020).
  • [20] B. H. He, H. A. Witek, J. Langner, R. Podeszwa, Can the John-Sachs theory of Kekulé structures be extended to enumerate Clar covers of benzenoids?, MATCH Commun. Math. Comput. Chem. (to be submitted) (2020).
  • [21] H. A. Witek, J. Langner, Clar covers of overlapping benzenoids: Case of two identically-oriented parallelograms, Symmetry 12 (2020) 1599 (20 pages).
  • [22] H. A. Witek, J. Langner, R. Podeszwa, Closed-form formulas for Zhang-Zhang polynomials of hexagonal graphene flakes O(k,m,n) with k, m=1-8 and arbitrary n, MATCH Commun. Math. Comput. Chem. (to be submitted) (2020).
  • [23] J. Langner, H. A. Witek, G. Moś, Zhang–Zhang polynomials of multiple zigzag chains, MATCH Commun. Math. Comput. Chem. 80(1) (2018) 245–265.
  • [24] Maple 16. Maplesoft, a division of Waterloo Maple Inc., Waterloo, Ontario. Maple is a trademark of Waterloo Maple Inc.