Unit interval parking functions and the -Fubini numbers
Abstract.
We recall that unit interval parking functions of length are a subset of parking functions in which every car parks in its preference or in the spot after its preference, and Fubini rankings of length are rankings of competitors allowing for ties. We present an independent proof of a result of Hadaway, which establishes that unit interval parking functions and Fubini rankings are in bijection. We also prove that the cardinality of these sets are given by Fubini numbers. In addition, we give a complete characterization of unit interval parking functions by determining when a rearrangement of a unit interval parking function is again a unit interval parking function. This yields an identity for the Fubini numbers as a sum of multinomials over compositions. Moreover, we introduce a generalization of Fubini rankings, which we call the -Fubini rankings of length . We show that this set is in bijection with unit interval parking functions of length where the first cars have distinct preferences. We conclude by establishing that these sets are enumerated by the -Fubini numbers.
Key words and phrases:
ennumerative combinatorics, parking function, fubini number, fubini ranking1991 Mathematics Subject Classification:
Primary: 05A05; Secondary 05A191. Introduction
Throughout we let and . In this work, we are concerned with studying and enumerating families of -tuples in satisfying certain properties and providing bijections between them. Our main objects of study are unit interval parking functions and Fubini rankings. We begin by defining each as subsets of and recalling some known results from the literature.
Parking functions, which were introduced by Konheim and Weiss [9] in the context of hashing problems, can be defined as -tuples such that at least entries of are at most , for all . We denote the set of parking functions of length by , and it is well-known that , [13].
We consider a special subset of parking functions called unit interval parking functions, as defined by Hadaway in [8]. In this case, one considers a queue of cars entering a one-way street with parking spots labeled increasingly from to . The preferences of the cars are provided in a preference list , where car prefers parking spot , for all . Car drives to its preference and if that parking spot is available it parks there. Otherwise, it attempts to park in spot . If that spot is available, then parking is successful. Otherwise, car fails to park. If all cars are able to park using this unit interval parking rule, then we say that is a unit interval parking function of length . For example, is a unit interval parking function while the rearrangement is not. We let denote the set of unit interval parking functions of length .
Now consider the set of possible rankings for how competitors can rank in a competition allowing ties. In this setting, no ranks can be skipped, whenever two competitors tie for a rank they “cover” that rank and the next, and similar whenever more than two competitors tie they cover. For example, is one such ranking, while is not. These rankings were studied by Gross [7], who showed that the number of such rankings is given by a Fubini number111The numbers were named after Fubini by Comtet in [4]. We recall that the Fubini numbers, also known as the ordered Bell numbers[12, A000670], are defined by
(1) |
Hadaway called these rankings “Fubini rankings” and established a bijection between them and unit interval parking functions [8, Theorem 5.12]. Her bijection depended on the content of a parking function, i.e. the multiset describing the values appearing in the tuple.
Motivated by the results of Hadaway [8], in this work we give a direct bijection between unit interval parking functions and Fubini rankings, which highlights their “block structure” (Definition 2.7). The block structure of these combinatorial objects plays a key role in our answer of Hadaway’s question of which rearrangements of a unit interval parking function is again a unit interval parking function. Moreover, this helped us uncover a new formula for the Fubini numbers. The block structure also allows us to fully generalize the result to a new family of combinatorial objects which we call -Fubini rankings.
This article is organized as follows. In Section 2, we study unit interval parking functions and we:
-
(1)
Give a complete characterization of unit interval parking functions by proving which rearrangements of unit interval parking functions are again unit interval parking functions (Theorem 2.9).
-
(2)
Establish a bijection between the set of unit interval parking functions of length and Fubini rankings of length (Theorem 2.5). This implies immediately unit interval parking functions of length are enumerated by the Fubini numbers, Equation (1). As a consequence, we give a new formula for the Fubini numbers (Corollary 2.13):
where denotes that c is a composition of with parts.
In Section 3, we introduce the -Fubini rankings of length (Definition 3.2). Our main results establish that they are enumerated by the -Fubini numbers. We recall that the -Stirling numbers of second kind222The -Stirling numbers of the second kind are [12, A143494]. are defined by
and the -Fubini numbers333Note that the -Fubini numbers are the Fubini numbers defined in (1). are a generalization of the Fubini numbers defined in [1, 11] by
- (3)
We end in Section 4 detailing some directions for future research.
2. Bijection between unit interval parking functions and Fubini rankings
In this section, we establish the bijection between and the set of Fubini rankings of length , denoted by (Theorem 2.5). We remark that this result was originally established by Hadaway in [8]. We begin by giving a formal definition of Fubini rankings.
Definition 2.1.
The -tuple is a Fubini ranking of length if the following holds: For all , if entries of are equal to , then the next largest value in is . Denote the set of Fubini rankings of length by .
Note that is a Fubini ranking, while is not. The following result is due to Cayley [2].
Proposition 2.2.
For , .
We remark that any rearrangement of a Fubini ranking is again a Fubini ranking, as a Fubini ranking only depends on the ranks and not who places in which rank. Recall that this is also true for parking functions as any rearrangements of a parking function is also a parking function. However, this is not true for unit interval parking functions. For example, is a unit interval parking function while is not. We soon address the question of when a rearrangement of a unit interval parking function is again a unit interval parking function (Theorem 2.9). Before doing so, we consider Fubini rankings as parking preferences and establish that Fubini rankings are always parking functions.
Lemma 2.3.
For all , .
Proof.
Suppose that is a Fubini ranking. Then by our previous remark we know that any rearrangement is also a Fubini ranking. Thus consider the weakly increasing rearrangement of which we call which satisfies that for all . To show that it suffices to check that for all . By Definition 2.1, there exists values , thus, values for all . Then, again by definition, the next entry is , which satisfies . Iterating this process, for if there are ties then or there is not tie and the value at is a new rank, hence . Thus, , for all , and . ∎
Observation 2.4.
Consider as a parking preference. Observe that cars at indices corresponding to a repeated value in do not interact with any car such that . Thus, if has parking preference , i.e. , then parks in spot , where denotes the number of occurrences of in up to car .
Next we provide an independent proof of the following result [8, Theorem 5.12].
Theorem 2.5.
If , then the sets and are in bijection, and .
To make our bijection precise we begin by defining the displacement of a parking function: Given a parking function , if car parks in spot , then is the displacement of car , and the sum is called the total displacement of . We note that there are interesting results related to the study of parking functions and their displacement. This includes a bijection that takes the total displacement of parking functions to the number of inversions of labeled trees [10]. In our work, displacement is utilized as a way to test whether a parking function is a unit interval parking function. Namely, we know that a parking functions is a unit interval parking function if and only if each car has displacement at most 1. In what follows, we say that whenever the displacement of a car is equal to zero, then the car is lucky. Lucky cars and tree inversions were studied in [6, 15].
To prove Theorem 2.5 we define a bijection from Fubini rankings (which we know are parking functions by Lemma 2.3) to unit interval parking functions that modifies the entries of the Fubini ranking. This map will do this in such a way so as to ensure that each car still parks in the same space, but is displaced at most one, which makes it a unit interval parking function.
Lemma 2.6.
Given , define by , where
The map is well defined.
Proof.
We show that is a unit interval parking function whenever is a Fubini ranking. To do so we induct on , the number of distinct values in .
If , then and . As a parking function, parks the cars in order , and they all park within one spot of their preference. Thus a unit interval parking function.
Suppose that if has distinct values, then is a unit interval parking function. Now consider having distinct values.
Let be the set of indices where the smallest distinct values of occur, and let be the set of indices where the max value of occurs. Note that by assumption, the entries in indexed by park in the first parking spots, and they do so by parking at most one away from their preference.
Now consider the entries in indexed by , which contain the values . Since is a Fubini ranking, we know that the next available rank is . Moreover note that , as otherwise the rank would be unearned and would not have been a Fubini ranking. To finish building , we need only consider the values in index set . By definition, those values in were all and in they have now been replaced with the values , appearing in this relative order at the indices in . We conclude by now noting that under the cars indexed by park in spots . Thus, the cars indexed by parking under are displaced at most one from their preference, as desired. ∎
In [8, Lemma 5.6] Hadaway established that if , then a value may appear at most twice. Below, we characterize when a rearrangement of a unit interval parking function is again a unit interval parking function.
To make our approach precise, we begin with the following general definition on parking functions, which in Theorem 2.9 below we specialize to unit interval parking functions.
Definition 2.7.
Given , let be the weakly increasing rearrangement of . The partition of as a concatenation denoted , where begins at (and includes) the th entry satisfying , is the block structure of . Each is called a block of .
For example, if , then and .
Observation 2.8.
Note that if is a unit interval parking function with block structure , then as cars in each block park, they park without interacting with cars in other blocks. To make this precise we let denote the length of for all . Then cars in park and occupy spots through . By definition of car has preference , and hence parks precisely in spot . As all other elements in are in weakly increasing order the remaining cars of park in spots . This shows that the cars in and park without affecting each other and this fact extends to all with .
Lastly, we note that each block of corresponds to a block of repeated values in a Fubini ranking, as described in Observation 2.4.
Theorem 2.9.
Given , let be its weakly increasing rearrangement and be the block structure of (as in Definition 2.7).
-
(1)
There are
(2) possible rearrangements of such that is still a unit interval parking function.
-
(2)
A rearrangement of is in if and only if the entries in respect the relative order of the entries in each of the blocks .
Proof.
To establish Part (1) we prove the following claims by inducting on the number of blocks in .
-
•
The weakly increasing rearrangement of an element in is also in .
-
•
If is a weakly increasing block of , then those elements appear in weakly increasing order in .
-
•
Any two blocks and are independent of each other, i.e. as sets for all .
-
•
The formula in (2) holds.
Suppose . Then is such that , since it is a parking function, and for by assumption. Thus, , which implies . That is and car 2 parks in spot 2 and is displaced 1 spot from its preference. For car 3 to have displacement at most 1 we must have since spots 1 and 2 are occupied so that , i.e. . Continuing in this fashion establishes . Next we establish that by supposing that is out of weakly increasing order to obtain a contradiction.
Let be the smallest index such that and there exists some satisfying . Let be the largest such . Parking under , cars park in spots , respectively. Car with preference parks in spot leaving spots unoccupied after it parks. Note that of the cars numbered , those with preferences in the set park without displacement and those with preference smaller that get displaced one unit. Also, one of the cars with preference smaller than has parked in spot . Further, any car with preference larger than parks with a displacement at most one. This is implied because consists of the numbers so that all numbers not equal to one appear exactly once in . This then ensures that spots are occupied by cars arriving and parking before car . Thus, the displacement of car is at best , which contradicts that . Therefore, the unique unit interval parking function consisting of a single block is , and it is in weakly increasing order. Furthermore, there are possible rearrangements that produce a unit parking function.
Suppose that if is such that has distinct blocks, then those blocks appear in as weakly increasing strings, and there are , ways to reorder so that the result is again in .
Now consider having distinct blocks in . We may consider the block structure of as the concatenation , where the block structure of is given by , and the block structure of is given by . As in Observation 2.8 we may now consider both and as unit parking functions on their own, also noting that , otherwise we would not have a parking function.
By induction, , and there are ways to rewrite it as a unit parking function, by induction hypothesis. Separate from this, has the form of an element in , and there is one way to arrange these elements. Because fills parking spots , it does not interfere with being a unit parking function, and thus and can be parked “without interactions.” That is, cars can be given preferences from a valid rearrangement of at the same time as cars are given preferences from . Then if is a unit parking function produced by rearranging , there are
ways that we could intermix the preferences from and . By induction, there are ways to rearrange as a unit parking function, thus there are
ways to rewrite as a unit parking function, as desired.
To prove Part (2) first note that the above proof of Part (1) implies that if respects the relative orders of the entries in the blocks, then . It suffices to prove the reverse direction, that given a unit interval parking function, then it respects the relative order of the entries in each block. To show this, note that the set of cars with preferences within a block of park independently of each other. We know that every block of must be a unit interval parking function in its own right, if is a unit interval parking function. From the base case of Part (1), we established that a single block is a unit interval parking function if and only if its entries are in weakly increasing order. Thus, a unit interval parking function must have all of the values within its blocks appearing in weakly increasing order, as claimed. ∎
Remark 2.10.
We are ready to introduce the inverse to the map in Lemma 2.6.
Lemma 2.11.
Given , let be its weakly increasing rearrangement and be the block structure of (as in Definition 2.7). Define by , where, for all ,
The map is well defined.
Before establishing Lemma 2.11 we provide an example to illustrate the map .
Example 2.12.
Let and consider . We leave it to the reader to verify that . The weakly increasing rearrangement and the block structure of , denoted , are as follows:
For each , is the smallest element in the block containing the value ; that is, . One can verify that is a Fubini ranking.
Proof of Lemma 2.11.
Let and let be the weakly increasing rearrangement of .
Let be the set of indices satisfying and let . Then the block structure of can be partitioned as the concatenation , where a new block begins at (and includes) each satisfying .
Because is a unit interval parking function, and thus is a parking function, we note that for all . This means that if , and is the minimum element of a new block, this is the first occurrence of the number in . Therefore, there are elements in less than , all appearing in the blocks before whichever block contains . Let . This integer, , may then be partitioned as follows:
Therefore, in general, we have that
We also note that the elements and then have the relationship . Moreover, by definition of , there are exactly copies of the value in . Also, note that by definition begins with 1, and so , and hence there exists a in . These facts together imply that is a Fubini ranking. ∎
The following result gives a formula for the Fubini numbers, which we did not find in the literature.
Corollary 2.13.
If and denotes a composition of with parts, then
Proof.
We are now ready to prove our main result.
Proof of Theorem 2.5.
It suffices to show that and are inverses of each other.
First, we prove that for every , . Suppose has distinct values, .
For each , let be the set of indices containing the value in . Since is a Fubini ranking, we have that for all . (In particular, .) For each , takes the values at index set in and replaces them by the values at index set (in that relative order) in .
To apply to , we first rearrange into weakly increasing order, which is given by
To partition the block structure of we begin by noting that, for all , since and the first appears in index , the first instance of begins a new block. Thus
where, for all , we have that
(3) |
Now observe that the entries in are defined by
Claim: for all .
To prove the claim we consider the values in the index set . In , these are ’s. In , these are the values in which are (still occurring at the indices ). In for any , the value is . So for every , which is precisely the value of at every . Since this holds for all and since , we have established that (showing the lists agree at every entry). Hence .
For the reverse composition, we begin with a unit interval parking function and let be the weakly increasing rearrangement of and we partition the block structure of as
where a new block begins and includes each satisfying . For each , let . Then use Lemma 2.11 to find where . Note is a Fubini ranking. Now consider .
Claim: for all .
To prove the claim we consider the values and where for an arbitrary . Fix and let where . Let denote the substring of made up of the values at the indices in . Note that by definition of , for all . Let denote the substring of made up of the values at the indices in . By definition of ,
By construction/definition and for all . By Theorem 2.9, these values appear in in the exact same relative order. Thus, for all . ∎
3. Bijection between unit interval parking functions starting with distinct values and -Fubini rankings
We begin this section by defining -Fubini numbers and show that they enumerate -Fubini rankings. We also establish that the -Fubini rankings are in bijection with the set of unit interval parking functions, which begin with distinct values.
Definition 3.1.
The -Fubini numbers444The -Fubini numbers are defined in [12, A232472]. are given by
where denotes the -Stirling numbers of the second kind555The -Stirling numbers are defined in [12, A008277]., which enumerate set partitions of length into blocks, in which the first values appear in distinct blocks.
For ease of reference, we provide Table 1, where we give the values of these numbers for , , and .
1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | |
1 | 1 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
2 | 3 | 2 | 0 | 0 | 0 | 0 | 0 | 0 |
3 | 13 | 10 | 6 | 0 | 0 | 0 | 0 | 0 |
4 | 75 | 62 | 42 | 24 | 0 | 0 | 0 | 0 |
5 | 541 | 466 | 342 | 216 | 120 | 0 | 0 | 0 |
6 | 4683 | 4142 | 3210 | 2184 | 1320 | 720 | 0 | 0 |
7 | 47293 | 42610 | 34326 | 24696 | 15960 | 9360 | 5040 | 0 |
8 | 545835 | 498542 | 413322 | 310344 | 211560 | 131760 | 75600 | 40320 |
Definition 3.2.
An -Fubini ranking of length is a Fubini ranking of length whose first values are distinct. More precisely, is an -Fubini ranking if . We let denote the set of -Fubini rankings of length .
Example 3.3.
There are ten -Fubini rankings of length three with the first two values being distinct. These are: . Note that this is precisely the value in Table 1 with and .
Note that when , the -Fubini rankings allow the repetition of any of the values, and thus we recover the definition of Fubini rankings given by Hadaway [8] and which we studied in Section 2. Hence, the -Fubini rankings are enumerated by the Fubini numbers. Moreover, the -Fubini rankings are nested, meaning that
where denotes the set of permutations on the set .
In what follows we let denote the set of unit interval parking functions of length in which the first values are distinct. As with -Fubini rankings, the sets are also nested:
Even though the sets share similar properties, such as this nesting, we now provide an example which illustrates that the sets and are not the same.
Example 3.4.
Note is in but not in , since car would not park within a unit interval. Note is in and not in , because because the tie at rank 4 would disallow rank 5 from appearing.
We are now ready to state and prove our main result of this section.
Theorem 3.5.
For and , if , then the sets and are in bijection.
Proof.
Given a in which the first values are distinct, we use the bijection in Theorem 2.5 to find a unique . We know that each of the first cars will park in its preferred spot, so each of those values will be starting a new block and will thus be mapped to itself in the Fubini ranking. So the Fubini ranking will start with the same distinct values as the unit interval parking function. Thus, . ∎
Next we enumerate the elements of .
Theorem 3.6.
If and , then .
Proof.
Let be weakly increasing. As , by Theorem 2.9 we know the structure of , meaning that there exists , with , indices of such that . These values always begin a block . By Theorem 2.9, we know that we can construct all elements of by permuting the blocks while keeping the relative order of the elements of each block. The number of ways to do this, for a fixed , is given by
where (as in Definition 3.1) the definition of the -Stirling number ensures that the first values appear in distinct blocks. The result follows by taking the sum over all possible . Therefore,
Corollary 3.7.
If and , then .
4. Future Work
To begin, we wonder if it is possible to construct a simple proof of Part 2 of Theorem 2.5, that does not rely on Part 1, as that would eliminate some of the complexity in our argument. Moreover, there is much to be discovered about -Fubini rankings and we now provide some directions for further study:
-
(1)
We remark that the intersection between Fubini rankings and unit interval parking functions is non-trivial. In fact, computationally the sequence for is given by [12, A080599] whose terms for are:
Stanley notes that this sequence gives the number of intervals in the weak (Bruhat) order of that are Boolean algebras. A new bijective proof has now appeared [5].
-
(2)
One could extend the above to the case of -Fubini rankings with . Namely, enumerate the cardinalities of the set , where we fix one of the parameters while varying the other.
-
(3)
The study of statistics of permutations is a well-researched mathematical area. Schumacher has results related to ascents, descents, and ties for parking functions [14] and it would be of interest to study statistics for unit interval parking functions and Fubini rankings, as well as their generalizations and . We hope to follow up this paper with results in this direction.
References
- [1] Leonard Carlitz. Weighted stirling numbers of the first and second kind–ii. Fibonacci Quart, 18(3):242–257, 1980.
- [2] Arthur Cayley. On the analytical forms called Trees, with application to the theory of chemical combinations, volume 9 of Cambridge Library Collection - Mathematics, page 427–460. Cambridge University Press, 2009.
- [3] Lucas Chaves Meyles, Pamela E. Harris, Richter Jordaan, Gordon Rojas Kirby, Sam Sehayek, and Ethan Spingarn. Unit-interval parking functions and the permutohedron, 2023. Arxiv:2305.15554.
- [4] Louis Comtet. Advanced combinatorics. D. Reidel Publishing Co., Dordrecht, enlarged edition, 1974. The art of finite and infinite expansions.
- [5] Jennifer Elder, Pamela E. Harris, Jan Kretschmann, and J. Carlos Martínez Mori. Boolean intervals in the weak order of , 2023. Arxiv:2201.12887, accepted at Journal of Combinatorics.
- [6] Ira M. Gessel and Seunghyun Seo. A refinement of Cayley’s formula for trees. Electron. J. Combin., 11(2):Research Paper 27, 23, 2004/06.
- [7] Oliver A. Gross. Preferential arrangements. Amer. Math. Monthly, 69:4–8, 1962.
- [8] Kimberly P. Hadaway. On combinatorical problems of generalized parking functions. Williams College, Honors Thesis, 2022.
- [9] Alan G. Konheim and Benjamin Weiss. An occupancy discipline and applications. SIAM Journal on Applied Mathematics, 14(6):1266–1274, 1966.
- [10] Germain Kreweras. Une famille de polynômes ayant plusieurs propriétés énumeratives. Period. Math. Hungar., 11(4):309–320, 1980.
- [11] István Mező. Periodicity of the last digits of some combinatorial sequences. Journal of Integer Sequences [electronic only], 17, 08 2013.
- [12] OEIS Foundation Inc. The On-Line Encyclopedia of Integer Sequences, 2023. Published electronically at http://oeis.org.
- [13] John Riordan. Ballots and trees. J. Combinatorial Theory, 6:408–411, 1969.
- [14] Paul R. F. Schumacher. Descents in parking functions. J. Integer Seq., 21(2):Art. 18.2.3, 8, 2018.
- [15] Seunghyun Seo and Heesung Shin. A generalized enumeration of labeled trees and reverse Prüfer algorithm. J. Combin. Theory Ser. A, 114(7):1357–1361, 2007.