A bijective toolkit for signed partitions William J. Keith Abstract. The recently formalized idea of signed partitions is examined with intent to expand the standard repertoire of mappings and statistics used in bijective proofs for ordinary partition identities. A new family of partitions is added to Schur’s Theorem and observations are made concerning the behavior of signed partitions of 0 in arithmetic progression. Mathematics Subject Classification (2000). Primary 05A17; Secondary 11P83. Keywords. partitions, signed partitions, partitions of zero, complement, Ferrers diagram. 1. Introduction In a recent article [2], Andrews examined the notion of partitions in which we allow some parts to be negative, calling these signed partitions and producing extensions of some classical theorems of the subject. His proofs were all analytic: short generating function manipulations that work for negative exponents. This paper begins exploring the toolkit of bijections and combinatorial statistics for its use with negative parts. We start by reproving Andrews’ theorems bijectively, usually with refinements, and then look further abroad. For example, we can apply bijective techniques to add an intermediary class of signed partitions to a theorem of Schur, giving a short bijective proof of several of its classical clauses: Theorem 1. Partitions of n into distinct nonmultiples of 3 are equinumerous with signed partitions of n in which all positive parts are divisible by 3, each at least 3 times the number of positive parts, with negative parts not divisible by 3, all less than 3 times the number of positive parts and having distinct 3-weights, but −(3(ki + 1) ± 1) both appearing only if 3(ki + 1) does. Formally, we can regard a signed partition of an integer n as a finite sequence (λj , . . . , λ1 , λ−1 , . . . , λ−k ) where λi ∈ Z \ {0}, λj ≥ · · · ≥ λ1 > 0 > λ−1 ≥ · · · ≥ λ−k , and λj + · · · + λ−k = n. (There are many equivalent ways of viewing such partitions, each with their own context and utility. We shall encounter several in this article.) 2 William J. Keith Considered, as Andrews did, in their context from the work of Euler, these new partitions are quite old objects. However, perhaps due to analytic concerns for the convergence of the generating function – there is, after all, an infinite set of such partitions for any n, if we do not restrict the kinds of parts allowed – little research has been done on them in the intervening centuries. This may have been a loss to the subject. As Andrews points out, several classical partition identities can be cast in this light. Perhaps more usefully, once our eyes have been opened signed partitions by other names appear in various places in the literature. Garrett [7] discusses Laurent-polynomial generalizations of the Rogers-Ramanujan identities and others from Slater’s list, using lattice paths that start below the x-axis in a manner that cries out for combinatorial interpretation in terms of signed partitions. In a representationtheoretic context, they can be interpreted as the signatures of analytic representations of GL(n) (as in [12]). Study may thus be more fruitful than previously imagined. In Section 2 we reprove bijectively a related group of four theorems from [2], refining some of them in doing so, and from there expand our search to other identities. In Section 3 we take note of the techniques we are using and define them for signed partitions in general. We demonstrate some utility by adding a morsel to a theorem of Schur to produce a short bijective proof. Possible adaptations of the Ferrers diagram for partitions are discussed, along with uses of the diagram such as conjugation, profiles, and complementation. We close with some open questions. The standard notation in q-basic hypergeometric series is used throughout: (a)n = (a; q)n = (1 − a)(1 − aq)(1 − aq 2 ) . . . (1 − aq n−1 ) (1) (q)n = (q; q)n (2) (a)∞ = (a; q)∞ = lim (a; q)n (3) n→∞ 2. Four Theorems on Odd and Distinct Parts The four theorems of interest from [2], appearing there as Theorems 7 through 10, are as follows: Theorem 2. Let E(m, n) denote the number of ordinary partitions of n with exactly m even parts, none of which is repeated. Let S(m, n) denote the number of signed partitions of n in which each positive part is larger than the number of positive parts, the negative parts are all distinct, each not exceeding the number of positive parts, and m designates the excess of the number of positive parts over the number of negative parts. Then E(m, n) = S(m, n). Theorem 3. Let G1 (n) denote the number of signed partitions of n in which each positive part is even and at least twice the number of positive parts, and the negative parts are distinct and odd with each smaller than twice the number of positive parts. Let G2 (n) A bijective toolkit for signed partitions 3 denote the number of ordinary partitions of n in which each part is ≡ 1, 4 or 7 (mod 8). Then G1 (n) = G2 (n). Theorem 4. Let S1 (n) denote the number of signed partitions of n with an odd number (say j) of positive parts, each positive part ≥ (j − 1)/2, with distinct negative parts each odd and < j. Let S2 (n) denote the number of ordinary partitions of n into parts not divisible by 4. Then S1 (n) = S2 (n). Theorem 5. Let R1 (n) denote the number of signed partitions of n with an odd number (say j) of positive parts, each positive part ≥ (j − 1)/2 with distinct negative parts each ≤ (j − 1)/2. Let R2 (n) denote the number of ordinary partitions of n into parts that are either odd or ≡ ±2 (mod 10). Then R1 (n) = R2 (n). There are some clear commonalities among all these theorems. Each deals with partitions into odd parts, and partitions into distinct parts, two sets of partitions with several commonly known bijections relating them. In each set of signed partitions we are equipped with a number of positive parts each exceeding a given minimum, and negative parts distinct and restricted in size and therefore number. These sizes are small enough that, after at most minor manipulation, the negative part of the partition is dominated by the guaranteed portion of the positive part. Visually, it “fits within” the guaranteed positive part, and immediately the procedure suggested to us is to subtract the negative parts from the guaranteed positive parts, leaving a pair of ordinary partitions. To construct a bijective proof of the theorems, it remains then merely to combine the pair into a single partition in a way that records the identities of the original pair. Proof of Theorem 2. This theorem is provided with an extra statistic our bijection must preserve – that the number of even parts in the ordinary partition is the same as the excess of the number of positive parts over negative parts in the associated signed partition. This statistic will appear again in our refinements of the other theorems, and the additional structure makes it a good theorem to display the underlying intuition in detail. Begin with an element of the set of signed partitions counted by S(m, n). Suppose that the number of positive parts is j. In an extension of the usual Ferrers diagram, in which we view the parts of an ordinary partition as rows of lattice points in the fourth quadrant of length equal to each part, we can view a signed partition as rows of lattice points first and third quadrants, thus: In Figure 1 we have a diagram of an element of S(3, 70) with 8 positive and 5 negative parts, the signed partition λ = (17, 15, 12, 11, 10, 10, 9, 9, −1, −3, −4, −7, −8). We have highlighted the portion of each positive part equal to the number j of positive parts (which in the future, following Andrews, we will call the + number, and similarly k the − number). We are guaranteed that this square and the column following it are fully occupied. There is also highlighted a square of equivalent size which by the definition of S(m, n) must contain the negative parts. Indeed, since the negative parts must be distinct, if bottom-justified in the square they would fit below its diagonal. We would like to transform this signed partition into an ordinary partition with 3 distinct even parts, and it seems reasonable that we should do so in some fashion clearly starting from the number of positive parts, and reducing by the number of negative parts. 4 William J. Keith F IGURE 1. An element of S(m, n). It is suggestive at this point to observe the final step in the generating-function proof of the theorem, which employs the identity qj 2 +j j z (1 + 1 1 1 )(1 + 2 ) . . . (1 + j ) = (−zq)j q j(j+1)/2 . zq zq zq The combinatorial interpretation is clear: we take the j 2 + j required minimum rectangle of the positive partition, and split it along its diagonal into two triangles of base j. One we leave in the positive portion, where as a standard construction the addition of j to the largest part, j − 1 to the next largest part, j − 2 to the next largest part, etc., to a partition with j or fewer parts results in a partition with exactly j parts all distinct. From the remaining list of positive parts {1, 2, . . . , j} we remove exactly those parts which appear in the negative part of the partition, leaving a second partition into exactly j − k distinct positive parts. We now have a pair of partitions (α, β), α with exactly j distinct parts and β with exactly j − k distinct parts. In figure 2 we illustrate this transformation performed on λ, resulting in the pair α = (16, 13, 9, 7, 5, 4, 2, 1), β = (6, 5, 2). It remains now to combine these two partitions in a way that recoverably preserves their identities, and produces exactly one even part for every part in β. It is clear that we can recover the original signed partition from this pair, and since we no longer deal here with negative parts we will summarize the rest of the bijection. Partitions of n into distinct parts are equinumerous with partitions into odd parts, and there are several bijections that realize this (see the survey of Pak [9]); we use the bijection of J. J. Sylvester (which may be found in many places, including [13]), which when applied to a partition with j parts produces a partition into odd parts in which at least d 2j e parts are of size at least 2b 2j c + 1. Call this γ. Any even parts in β we append (in the appropriate order) to γ; these will all be of size less than 2b 2j c + 1. Any odd part 2m + 1 of β is added to γ in a process that can be visualized as writing the part as 22 . . . 21 and “laying this along” γ: we add 2 to γ1 through γm , and 1 to γm+1 , this last addition producing a new even part. Since A bijective toolkit for signed partitions 5 m ≤ d 2j e−1, the 1 is added to an odd part of size at least 2b 2j c+1, producing an even part larger than this amount. Furthermore, all even parts produced by the process are distinct. This process is illustrated in Figure 3. Sylvester’s bijection on α produces γ = (19, 13, 11, 9, 3, 1, 1), with 4 parts larger than 8. From β we append parts 6 and 2, resulting in the partition (19,13,11,9,6,3,2,1,1); finally, we add the 5 from β by adding 2 to the first part, 2 to the second part, and 1 to the third part. Our final partition is (21, 15, 12, 9, 6, 3, 2, 1, 1). To reverse the bijection we subtract 1 from the largest even part of the partition and 2 from all larger odd parts to produce an odd part 2m + 1 of β, continuing until the remaining even parts are smaller than half the number t of odd parts larger than 2t; the remaining even parts are the even parts of β. We then reconstruct the signed partition. F IGURE 2. After subtraction. ¤ Theorem 4 is closely related to Theorem 2, sinceS the set S2 (n), partitions of n with no part divisible by 4, is equinumerous with the set m E(m, n), partitions of n with distinct even parts. To see this, we simply take from largest to smallest the even parts in an element λ of S2 (n), all of which will be twice an odd number ki ; then (k1 , . . . ) is a partition into odd parts, which may be transformed into a partition into distinct parts. If there are m of these, then the partition formed from collecting and properly ordering twice F IGURE 3. The final placement of parts. 6 William J. Keith the resultant distinct parts, and the odd parts of λ, is a new partition of n with distinct even parts. With this in mind, we can prove Theorem 4 bijectively by showing a correspondence between E(m, n) and exhaustive disjoint subsets of S1 (n). With a sufficiently interesting characterization of the latter subsets, this proof is then a refinement of the original theorem. This we do: define S1 (m, n) to be elements of S1 (n) with exactly m negative parts. Then S1 (m, n) is equinumerous with E(m, n). Proof of Theorem 4, refined. For α an element of S1 (m, n) with + number j, we set l = (j − 1)/2. Write α = (β2l+1 + l, . . . , β1 + l, α−1 , . . . , α−m ) and let β be the partition (β2l+1 , . . . , β1 ), β 0 the standard conjugation of β, and βo (resp., βe ) the partition consisting of the odd (resp. even) parts of β 0 . The 2l + 1 parts of size at least l in α provide the parts sufficient to dominate the negative parts, which are of size at most 2l + 1 and of which there are at most l. Once again we will construct a partition into odd parts and from it produce a controlled number of even parts; however, the process this time is a reflection S of the previous. Define for partitions λ and µ the partition π = λ µ to be the partition produced by collecting the parts of λ and µ and reordering according to size; define the partition π = λ + µ to have parts being the partwise sums of the parts of λ and µ, filling out the shorter of the two with parts 0 where necessary. Note that 2l + 1 parts of size l have the same cardinality as the triangular partition (2l, 2l − 1, . . . , 3, 2, 1). Form the new partition δ = φ ((2l, 2l − 1, . . . , 3, 2, 1) + β), where φ is Sylvester’s bijection as applied to a partition into distinct parts. It can be shown [10] 1 that the result of this rearrangement of the positive parts of α is the same as beginning with the partition (2l + 1, . . . , 2l + 1) S of l parts, constructing (2l + 1, . . . , 2l + 1) βo , and “laying flat” the parts in βe by twos, in a fashion analogous to the previous proof. Viewed either way, the result is a partition into odd parts. We construct the target partition in E(m, n) using a variant of a map combining pairs of partitions found in [4], for direction to which the author is grateful to Dr. Andrews. Starting with the last (that is, most negative, largest absolute value) negative part α−m = −(2k−m −1), subtract 2k−m −1 from the k−m -th part of δ, producing an even part which is reordered into place. Repeat for each part α−(m−1) through α−1 . The positive parts of δ are guaranteed to be large enough to accommodate the subtraction, and the even parts thus produced will all be distinct. There will be exactly one of them for every negative part in the original signed partition. An illustration of this process applied to the positive parts of such a partition is seen here in figures 4 through 6. To reverse the bijection, start with the largest even part, and add 2k + 1 where k is the unique value that will produce the k-th largest odd part in the partition. ¤ 1 The thesis of Stockhofe cited provides an excellent framework for understanding Sylvester’s bijection and generalizes it from odd and distinct parts, understood as modulus 2, to any modulus m. Written in German, it is available in English translation as an appendix of the doctoral thesis of the current author [8], who hopes to make it more widely available along with journal publication of the relevant portion of his thesis. A bijective toolkit for signed partitions 7 F IGURE 4. The signed partition α = (8, 6, 6, 5, 4, 3, 3, −1, −5). The gray rectangles are of size j = 2l + 1 by l, comprising the required parts of the positive portion of the partition and containing the possible negative parts (the striped region of length 2 is forbidden due to strict inequality). F IGURE 5. α with the positive parts rearranged into odd parts (reading the positive portion of the diagram by columns). F IGURE 6. The subtraction process for Theorem 4, with the diagram being read by columns. Combined with the previous proof, this proof defines a bijection between the sets of signed partitions S1 (m, n) and S(m, n). It might be instructive to produce a more direct map. 8 William J. Keith Pausing for a moment of analysis, let us observe the process which produced the even parts in each proof. In the first theorem, the number of even parts to be produced was equal to the excess of the number of positive parts over negative parts; we used negative parts to “cancel” elements from a list drawn directly from the positive portion, and each of the remaining positive parts was added to produce an even part. In the second theorem, the number of even parts to be produced was equal to the number of negative parts themselves, and the production was accomplished by direct subtraction from sufficiently large odd parts. If the subtraction of negative parts is to be a feature of bijective proofs for identities with signed partitions and one behavior or the other holds in the identity, this suggests we might look at the corresponding technique for proof. Bearing this in mind, let us turn to Theorem 3. As Andrews points out in [2], this theorem is a variation of the first Göllnitz–Gordon identity. That theorem is implied by a more general theorem ( [5], Theorem 7.11), for which the relevant specific case states that partitions into parts ≡ 1, 4, or 7 (mod 8) are equinumerous with partitions in the class D2,2 , defined as those partitions (b1 , . . . , bs ) for which bj − bj+1 ≥ 2 if bj is odd, bj − bj+1 > 2 if bj is even, and at most 1 part is 1 or 2. (The latter clause is, of course, implied by the former; it is retained for consistency with a more general statement discussed later.) Hand calculation of the two types of partition is easy for small n, and swiftly leads to the conjecture that signed partitions of type G1 (n) with excess of positive parts over negative parts m occur equally often with partitions in class D2,2 with exactly m even parts, or, that those with exactly m negative parts are equinumerous with those having m odd parts. The proof using the technique above is now almost instant. Proof of Theorem 3, refined. In a signed partition α from G1 (n), the required positive portion is of size 2j by j, or a square of 2s. Rearrange this square into a list of parts (2 · 1, 2 · 3, . . . , 2 · (2j − 1)), and add those parts above. We get a partition β into exactly j distinct even parts, say (β1 , . . . , βj ). For each negative part −(2k1 + 1), subtract 2 from β1 through βki and 1 from βki +1 , producing an odd part at the end. The process is illustrated in figure 7. The resulting partition is clearly of class D2,2 , and by reversing the prescribed recipe for subtracting parts, incrementing each even part by 1 and adding 2 to larger parts, we reverse the bijection. Each negative part produces an odd part in the ordinary partition, and the number of even parts is precisely the excess of the + number over the − number. ¤ (We could also understand this map in the following way: split the square into two triangular halves, subtracting 1 from α1 , 3 from α2 , etc., and 2j − 1 from αj , leaving a partition β into exactly j distinct odd parts, say (β1 , . . . , βj ). From the set of magnitudes subtracted, that is, {1, 3, . . . , 2j − 1}, remove any negative parts which appear in α. Suppose that the list remaining is γ = {2k1 + 1, . . . , 2km + 1}. Add these parts to β by “laying flat”: for each part 2ki + 1 of γ, add 2 to β1 through βki and 1 to βki +1 , producing an even part.) This proof is so simple that it lends itself to expansion. On the one hand, it would be nice to be able to prove the entire family of identities alluded to in the paragraph before the proof, which in general state that for 0 < i ≤ k partitions of n into parts 6≡ 2 (mod 4), and A bijective toolkit for signed partitions 9 F IGURE 7. The bijection for Theorem 3, applied to α = (14, 10, 10, 8, −1, −3, −7). The partition splits into the pair β = (13, 7, 5, 1), γ = (5). The pair combines to form the target partition (15, 9, 6, 1). 6≡ 0, ±2i − 1 (mod 4k), are equinumerous with partitions of the class Dk,i (n), partitions with odd parts, distinct, at most i − 1 parts of size 1 or 2, and differences among parts bj − bj−k+1 ≥ 2 if bj is odd and > 2 if bj is even. The first example would be the case k = i + 1 = 2, which relates to the second of the Göllnitz–Gordon identities. The same hypotheses define D2,1 as define D2,2 save that no parts can be 1 or 2. From our bijection, it is easy to define the class of signed partitions that gives rise to this set: they are simply those signed partitions with the same hypotheses as in the theorem, save that the even positive parts are strictly larger than twice the + number. The case for modulus k > 2 presents a different appearance. A more complete exploration we undertake in the next section, after proving the fourth of the theorems we are beginning with. The simplicity of this proof also makes it a good showcase for an illuminating principle: there can be more than one way to combine the positive and negative parts of a signed partition, each giving rise to different classes of partitions, or the same classes with different statistics preserved. For example, in the theorem above we could take the positive parts and conjugate their 2-modular diagram (that is, divide all parts by 2, conjugate the resulting partition, and multiply the conjugate by 2 again). This partition has at least j largest parts of size exactly 2j. From this partition we can reversibly perform a direct subtraction of the negative parts, removing −(2ki − 1) by subtracting 2ki − 1 from part ki , and reordering the resulting odd part into place. The resulting ordinary partition of n has distinct odd parts, of a number equal exactly to the number of negative parts in the original signed partition of n, and the class is further defined by the restriction that, when the largest odd part is 2m + 1, either all parts from {1, 3, . . . , 2m + 1} appear or 10 William J. Keith (exclusively of the previous) the largest part is even, and in the latter case appears at least a number of times equal the number of parts from {1, 3, . . . , 2m+1} not appearing. More such classes can be produced; all are immediately equinumerous. Before we head off in new directions, let us conclude our tour of Andrews’ paper with his Theorem 10, number 5 in our listing. It may be noticed that the positive parts of the signed partitions in this theorem are exactly the same as those of Theorem 4, while the negative parts would fit the hypotheses of that theorem if doubled and increased by 1. On the other hand, the Rogers-Ramanujan identities are consequences of the Göllnitz– Gordon theorem, so we might have expected a closer relationship between this set and the signed partitions of Theorem 3. It may be illuminating to consider this as we follow the proof. Proof of Theorem 5, refined. Once again, we employ the theorem mentioned by Andrews. The first Rogers-Ramanujan identity equates the number of partitions of n into parts ≡ 1, 4 (mod 5) with partitionsS of n into parts differing by at least 2. The set R2 (n) consists of those partitions α = αo αe into odd and even parts (notation as before), where the parts of αe are just twice parts ≡ 1, 4 (mod 5) and are thus equal in number to partitions of |αe | into even parts differing by at least 4. Bearing this identity in mind, we can prove the equivalence of the theorem’s two classes of partitions by showing equivalence between the signed partitions described, and partitions in which any even parts differ by at least 4. We would like, therefore, to produce ordinary partitions in which the even parts are under control. A handy way to do this would be to start with a partition entirely into odd parts, and so we manipulate the positive parts of the partition the same way we did for Theorem 4. If the + number is j = 2l + 1, and we call S the positive portion (l + β1 , . . . , l + β2l+1 ) with βi possibly zero, split β 0 into βo βe by odd and even parts, and note that S these parts ¡ are all ¢ at most 2l + 1 or 2l respectively. Construct γ = (2l + 1, . . . , 2l + 1) βo + 2 βe 0 /2 , where by the last term we mean adding βe by twos laying its 2-modular diagram along the previous partition. (See the earlier proof for an illustration of the process.) The result is a partition into odd parts. To construct even parts out of this partition we need to subtract odd parts, and are presently equipped with negative parts not odd but distinct. The solution is, of course, to employ Sylvester’s bijection once more, turning our distinct negative parts into odd ones. Since the largest of these is of size at most l, N.J. Fine proved analytically ([5], p. 26), and Christine Bessenrodt showed ([6]) that Sylvester’s bijection gives, that the resulting partitions into odd parts have their largest part plus twice the number of parts at most 2l + 1. Aesthetics suggest that our bijection will make maximum use of the leeway granted us in subtraction, and so we do. The map is the same one used in our proof of Theorem 4. Suppose that the transformed negative side of the partition is now δ = (δ1 , . . . , δs ). For each 1 ≤ i ≤ s, say δi = 2di + 1. Set 2ji − 1 = δi + 2(s − i). Starting in order with i = 1, subtract 2ji − 1 from part γji of our partition into positive odd parts, then reorder the parts. Next add a compensatory 2 to parts γ1 through γs−i . Since s − i + 1 ≤ l and γs−i+1 ≥ 2l + 1 > 2l − 1 ≥ 2ji − 1, the subtraction produces an even part of size at least 2 to begin with. Since ji is strictly decreasing as i approaches s, the compensatory A bijective toolkit for signed partitions 11 increment added to the larger parts of γ means after each step the next even part produced will differ from the previous by at least 4. We can recover γ and δ by reversing the procedure on such a partition, starting with the largest even part. We subtract 2 from the largest 0, 1, 2, etc. parts, then add the unique 2ji − 1 = 2di + 1 + 2(s − i) that produces from the largest remaining even part a new odd part which reorders into the ji -th place. There are numerous statistics preserved by Sylvester’s bijection which have been well-studied, and which can be translated into further descriptions of the behaviors of the even parts of the partitions produced here. For a concise summary, the reader is referred to [13]. ¤ Once again, a note on a related identity similar to our construction of the second Göllnitz–Gordon identity: the second Rogers-Ramanujan identity equates partitions in which differences between parts are at least 2, and the smallest part is greater than 1, with partitions into parts ≡ 2, 3 (mod 5). The related family can be constructed as in the above simply by insisting that no 2 ever be produced; for this we require that we do not simultaneously have both that 2j1 + 1 = 2d1 + 1 + 2(s − 1) = 2l − 1, and that γl = 2l + 1. In turn, these do occur if in the original signed partition the largest distinct negative part was of size −l and no part of size 2l appears in βe , i.e., the next-to-smallest positive part of α equals the smallest positive part. Thus, partitions of n in which the even parts are ≡ 4, 6 (mod 10) are equal in number to signed partitions of n in which there are an odd number j = 2l + 1 of positive parts, all of size at least l, negative parts are distinct and all ≤ l, and a negative part of size −l occurs only if the smallest positive part of α is distinct from the next-smallest. 3. New Directions While we have easily obtained a few refinements and some of the identities most closely related to our starting point, the field in which we are working now seems to have a much larger range of results in easy reach. 3.1. Partitions of (m, c)-type Let us begin by generalizing the modulus for the identities: each of the theorems described so far deals with partitions into odd and distinct parts, features apparent in the 2-modular diagram of a partition. For example, wherever a bijection calls for splitting a guaranteed amount of 2s into a triangle of odd parts, we can split a guaranteed amount of moduli m into parts congruent to c and m − c modulo m, and perform analogous manipulations. Such partitions are said to be of (m,c)-type, and are discussed in [13]. Theorem 2 does not seem, immediately, to yield to this sort of manipulation. Our signed partition there concerns distinct parts, not odd ones, and indeed, the bijection calls for carving our required positive parts into two triangles. On the other hand, Theorem 4 is much more adaptable to this viewpoint. The (m, c) version of that theorem (the refined version our bijection proved) is as follows: 12 William J. Keith Theorem 6. For 0 < c < m, let S1,(m,c) (k, n) denote the number of signed partitions of n with + number j ≡ c (mod m), with λi − λi+1 > 0 only if i ≡ 0, c (mod m), λj > |j|m , and k distinct negative parts ≡ −c (mod m), of size < j. (Equivalently, by taking conjugates, we require that the largest positive part be ≡ c (mod m), that it be repeated at least as often as its m-weight, and all positive parts are ≡ 0, c (mod m).) Let S2,(m,c) (k, n) be the number of ordinary partitions of n into parts ≡ 0, c (mod m) with exactly k parts divisible by m, all distinct. Then S1,(m,c) (k, n) = S2,(m,c) (k, n). Proof. Run through the bijection used to prove Theorem 4, using m-modular diagrams instead of 2-modular diagrams. The positive parts of a partition in S1,(m,c) (k, n), read by columns, are all of residue 0 or c modulo m. Replace the 2s in that bijection by m, and the 1s by c. We have a required square of ms of size j. Move positive parts divisible by m to lay flat on the top of the square, and subtract off negative parts −(tm − (m − c)) from the tth positive part thus produced. A partition in S2,(m,c) (k, n) results. ¤ We could, again, take the partition formed by the distinct parts of the target partition divisible by m, and perform Sylvester’s bijection upon their m-weights to obtain a list of parts not divisible by 2m, a class closer to that of the original theorem. We could also note, as we did before, that this class is of the same type as E(m, n) of Theorem 2. While it is not immediately obvious what the signed partitions should be, we should be able to reverse the direction of the bijection given in the proof of that theorem, replacing odd parts with parts ≡ c (mod m), to find out what such a class should be. Observe particularly Figure 3, remembering to read horizontally. The parts divisible by m would have been produced by either interleaving a part divisible by m, or “laying flat” a part ≡ m − c (mod m), either fitting to the side of or beneath the “m-modular Durfee square” – the visual square2 of size k in the m-modular diagram of a partition formed by moduli m when the first k parts are of size at least km – in a partition with parts all ≡ c (mod m). In the latter case the addition of c and m − c would produce a part divisible by m. Such a partition into parts ≡ c (mod m) would have arisen after applying the m-modular generalization Lm ◦ L1 of Sylvester’s bijection found in [10] (or the author’s English translation thereof, footnoted earlier) to a partition whose conjugate contained all parts ≡ 0, c (mod m) of consecutive m-weights up to the maximum appearing: i.e., parts c, m, c + m, 2m, c + 2m, . . . , c + (k − 1)m OR km. We are finally closing in on the starting signed partition. A partition of the type described above arises when we remove from a series of large, equal parts the (m, c) equivalent of a triangle: instead of alternating even and odd parts 1, 2, 3, . . . we remove parts m − c, m, 2m − c, 2m, 3m − c, . . . from the list of required parts, leaving parts (in reverse order) c, m, c + m, 2m, . . . . The sum of such parts in order is m, 2m, 3m, . . . . There are various ways to realize this list of required parts: we may require that the positive parts differ by at least m, the smallest being of size at least m, or we may describe it, in the spirit of our original theorem, in terms of the + number. With the latter approach, we have built the following generalization: 2 This is a simplified definition applicable to partitions in which the nonzero residues modulo m are all equal. A bijective toolkit for signed partitions 13 Theorem 7. Let E(m,c) (k, n) denote the number of ordinary partitions of n into parts ≡ 0, c (mod m), with exactly k parts divisible by m, none of which is repeated. Let S(m,c) (k, n) denote the number of signed partitions of n with + number j, k the excess of + number over − number, in which each positive part is at least as large as md 2j e, and if j is even the first 2j parts are at least m(d 2j e + 1); first differences λa − λa+1 are 0 among positive parts unless a ≡ 0, c (mod m), exception at least 1 at a = 2j if j even; and negative parts are ≡ 0, m − c (mod m), distinct, and not exceeding mj 2 (if j even) or m − c + mb 2j c (if j odd). Then E(m,c) (k, n) = S(m,c) (k, n). When seeing that this theorem really is a general case of Theorem 2, it should be noted that in that theorem m = 2, so c = 1 = m − c and all parts are automatically ≡ 0, c (mod m). Instead of requiring m = 2 be added to the first 2j parts of a partition with an even j, that theorem adds 1 to each of the required parts. The other theorems may be generalized similarly. Indeed, in some cases the (m, c) variants are so similar in structure to the original theorem that, as Zeng notes, the resulting generating function identities are algebraically equivalent to the originals with a substitution of variables. Additional richness can be ensured by considering instead, e.g., m-falling or m-rising partitions, in which the (least positive) residues of parts mod m are allowed to vary but must monotonically rise or fall. 3.2. Modulus 3 Heading in other directions, we quickly come across the realization that residues for larger moduli can be difficult to control when combining positive and negative parts of a signed partition – for m = 2 all such problems are swept under the rug since there is only one possible nonzero residue, which when added to itself yields m. The modulus m = 3, however, has a richer structure while still being controllable in convenient ways, some of which turn out to reflect interestingly on theorems present in the prior literature. Let us begin with a generalization of Theorem 3, in which rather than the (m, c) generalization we accept any residue on the negative parts: Theorem 8. The number of signed partitions α of n in which all positive parts are divisible by 3, each at least 3 times the + number, with negative parts not divisible by 3, all less than 3 times the + number and having distinct 3-weights, equals the number of ordinary partitions β of n in which all parts have different 3-weight, and the difference is greater than 1 if the larger of the two is divisible by 3. Proof. Note the slight recharacterization of distinctness required for the negative parts. When our modulus was 2, distinct odd parts were automatically of distinct 2-weight. Here we would disallow, say, parts -1 and -2, or -4 and -5. On the other hand, parts may differ by as little as 2, when we have, e.g., parts -2 and -4, so this condition is different from simply requiring that the negative parts differ by 3. Because the 3-weights of all negative parts differ, the proof used for Theorem 3 is adaptable in its entirety. We also obtain the refinements that the number of parts not divisible by 3, and divisible by 3, are equal to the number of negative parts and the excess of positive over negative parts, respectively. ¤ 14 William J. Keith A brief observation shows that this theorem holds with all instances of 3 replaced by any modulus. The 3-modular case is of particular interest, however, because the partitions involved are closely related to those in a theorem of Issai Schur. These were expanded upon by K. Alladi, Andrews, and finally Ae Ja Yee in a succession of papers we will follow (Andrews’ paper [3] is the start of this series, and is a good reference for the original theorem of Schur; the final paper in the sequence of interest here is [11]). These papers prove the identity of those sets with a number of others, combinatorially and by generating function manipulations. The original theorem is: Theorem 9 (Schur’s Theorem). Let A(n) denote the number of partitions of n into parts congruent to ±1 (mod 6). Let B(n) denote partitions of n into distinct nonmultiples of 3. Let D(n) denote partitions of n in which the differences between consecutive parts are at least 3, and strictly greater than 3 if 3 divides the larger part. Then A(n) = B(n) = D(n). Andrews and Alladi ([1], and see [3]) add a family of type C, partitions into odd parts appearing not more than twice, equivalent to those above. The paper [3] proves these equivalencies using short generating function manipulations, though it skips any proof at all that these three are equivalent to those of type D. Partitions of type B and C can be obtained bijectively from those of type Glaisher-style rewrites: map P A through P a partition in which a part 6k + 1, 5 appears bik 2i (resp. cik 3i ) times, bik = 0, 1 (resp. cik = 0, 1, 2) to that partition in which each part 2i (6k + 1, 5) appears bik times, i.e. partitions of type B (resp. each part 3i (6k + 1, 5) appears cik times, partitions of type C). The proof that these three classes are equivalent to type D is a much trickier affair from the viewpoint of q-series. However, we can add another equivalent family, of signed partitions, which can be rapidly mapped to those of types D and B, establishing this part of the equality. The partitions of D(n) are a subset of the ordinary partitions discussed in Theorem 8. The excess partitions are exactly those in which there are two consecutive parts (. . . , 3l + 4, 3l + 2, . . . ), which have different 3-weight but differ only by 2. Observe the signed partition from which such an ordinary partition arises in our proof: this occurs precisely when we subtracted from β negative parts of some size −(3ki + 2) and −(3(ki + 1) + 1), and there was not a column of 3s of length ki + 1 lying above the required square of 3s in the signed partition. If we select only those signed partitions in which such a column exists whenever we have successive negative parts of residue 2 followed immediately by 1, we obtain a class of signed partitions equinumerous with those of type D. Call such partitions of type F, and let F (n) enumerate those that partition n. Theorem 10. B(n) = F (n) = D(n). Proof. Partitions of type B can be produced from those of type F using a 3-modular variant of the proof technique we described just after proving Theorem 3. Take the square of required 3s and split it into a triangle of parts all 1 (mod 3), and an opposing triangle of parts all 2 (mod 3); list these. If the + number is j, all nonmultiples of 3 from 1 to 3j − 1 appear. We also have a list of multiples of 3 consisting of the excesses of the positive A bijective toolkit for signed partitions 15 parts over the required minimum. Read this list of excesses vertically to construct a list of positive parts divisible by 3, of size at most 3j. Remove from the list of nonmultiples of 3 any negative parts which appear. We remove at most 1 part of any given 3-weight, so the differences among the remaining parts are always ≤ 2 unless parts 3ki + 2 and 3(ki + 1) + 1 were removed; but in this case we defined our family of signed partitions so that one of the “excess” columns is exactly of size 3ki . We use the 3s of this column by adding 3 to the nonmultiple 3ki + 1 and each smaller part as far as we have 3s to add. The row of 3s thus produced will always be of length at least j + 1, and will not reach further than the last nonmultiple. The remaining excess columns of 3s from the positive part of our partition are all of size ≤ j; we add them unitwise in the same fashion, after completing the subtraction process above. We reverse the map by subtracting off rows of 3 unitwise until the rows get too long to have been excess 3s over the number of positive parts that could have produced our nonmultiples, given the subtractions that had to occur. Replace these missing nonmultiples, removing columns of 3s when necessary to add in −(3ki + 2) and −(3(ki + 1) + 1). Recombine the full list into a square and add the columns of 3s obtained in previous stages, listing the negative parts removed. ¤ In [3] Andrews adds, and in [11] Yee proves bijectively the equivalence of, a fifth class of partitions E(n) related to a family of orthogonal polynomials. The partitions of E(n) are those in which odd parts are distinct, even parts may appear at most twice, the difference between two parts is never 1, is 2 only if both are odd, and partitions with c c pairs of repeated even parts are counted witth weight (−1) . The partitions of type D are a subset of type E, and Yee produces a sign-reversing involution that fixes D and pairs off all other elements. It was hoped in this paper that the bijection might be simplified by considering the new class of signed partitions added to the group, but as of submission this has not been a successful search. A future investigator might find the challenge interesting. 3.3. Conjugation We have been using a new version of the useful old Ferrers diagram to illustrate signed partitions. Our presentation has served well in the construction of various bijections, and has the advantage that parts are listed vertically in descending order, including negatives. However, it may not be ideal for all purposes. One glaring feature it lacks is the phenomenally useful operation of conjugation, easily understood for the standard Ferrers diagram as reflection about the main diagonal in the plane. Any attempt to produce useful bijective tools for investigating signed partitions must consider some version of a generalization of this map. With the graphical presentation we have been using, reflection across either diagonal does not result in the diagram of a signed partition. The only linear transformation of the plane that does guarantee a new signed partition diagram is rotation about the origin by a half revolution. Numerically, this has the effect of mapping (λ1 , . . . , λj , λ−1 , . . . , λ−k ) → (−λ−k , . . . , −λ−1 , −λj , . . . , −λ1 ) i.e., reversing the order of parts and negating all. 16 William J. Keith Perhaps interestingly, this operation commutes with standard conjugation on ordinary partitions under the following map. Take an ordinary partition λ = (λ1 , . . . , λj ) and denote the standard conjugation of λ by λ0 . Suppose λ has (standard, or 1-modular) Durfee square of size s, i.e. s is the largest integer for which λs ≥ s. Then the Frobenius symbol of λ is given by ¶ µ λ1 − 1 λ2 − 2 . . . λ s − s . λ0 1 − 1 λ0 2 − 2 . . . λ 0 s − s s Each row is a partition (with possibly one zero part) into distinct parts. Standard conjugation on ordinary partitions switches the rows. If we add 1 to each entry and negate the bottom row, then list all elements of the Frobenius symbol as a signed partition, we obtain λ → (λ1 , λ2 − 1, . . . , λs − s + 1, −(λ0 s − s + 1), . . . , −(λ0 1 )) . This is a map between ordinary partitions of n with Durfee square of size s and the set of signed partitions in which the sum of the absolute value of all parts is n + s, with exactly s positive and s negative parts, each distinct. (Sign matters to distinctness; we allow parts j and −j.) Rotation of the resulting signed partition about the origin commutes with conjugation of the original ordinary partition. (A remark: if we added 12 instead of 1, the sum of the absolute values of the halfinteger parts thus produced would have been n regardless of s. The resulting vectors are apparently of interest in statistical physics concerning the distribution of energies among particle-antiparticle pairs; I am not expert in this field, but mention the connection for those whose curiosity might be piqued by related investigations.) Alternatively, we may seek a different diagram more suitable for a conjugation operation. Treating the positive and negative parts as separate partitions, and conjugating each, results in a new signed partition. This operation reduces to regular conjugation for ordinary partitions viewed as signed partitions with no negative parts. It is a linear transformation on the plane if we use a different diagram, thus: In the top of Figure 8, we have illustrated the signed partition (8, 7, 7, 6, 5, 4, 4, 3, −1, −3, −4, −7, −8) using a version of the Ferrers diagram in the fourth and second quadrants, with the negative parts in the second. While the parts are no longer visually arranged in descending order of magnitude, we can generalize conjugation visually. In the bottom left of the figure, we have reflected the diagram about the diagonal y = −x. The resulting partition is (8, 8, 8, 7, 5, 4, 3, 1, −1, −2, −2, −2, −3, −4, −4, −5), the partition wherein the positive and negative parts were each conjugated separately. This diagram and operation thus extend both the standard Ferrers diagram and the operation of conjugation to ordinary partitions as the special case of signed partitions, with no negative parts. In the bottom right of the figure, we employ the third transformation of the plane that preserves the double cone, reflection across y = x. It could also be understood as composing the previous two maps, rotation followed by separate conjugation, since these three maps are the ball game as concerns possible transformations. Separate conjugation A bijective toolkit for signed partitions F IGURE 8. Two versions of conjugation (8, 7, 7, 6, 5, 4, 4, 3, −1, −3, −4, −7, −8). 17 applied to is weight-preserving on signed partitions, but this operation, like rotation, exchanges a signed partition of n for one of −n. 3.3.1. Partitions of Zero. A somewhat astounding tangent arose at this point in investigations. In the usual study of partitions, there is only one partition of zero, the empty set. Any signed partition in which the positive part partitions n and the negative part partitions −n is a partition of zero, and it will remain so under the last reflection operation described above. 2 Call such a signed partition of height n; there are p(n) such partitions, p(n) being the number of ordinary partitions of n. Being a partition of zero is a property preserved under many more transformations of the plane than weight in general; for example, multiplying all parts by c preserves the property. An early stab at illustrating such partitions represented them as convex paths over the x axis: 18 William J. Keith F IGURE 9. Three partitions of 0, of height 11, illustrated as convex polygons on the axis. j X λi , j) (0, 0) → (λ1 , 1) → (λ1 + λ2 , 2) → · · · → ( i=1 j X λi − λ−1 , j + 1) → · · · → (0, j + k) . (4) →( i=1 These vary from “thinnest” at (n, −n) to “thickest” at (1, . . . , 1, −1, . . . , −1). The author became curious as to how the areas were distributed between their minimum of n and their maximum of n2 . In particular, it seemed natural to ask how many partitions there were of a given area at all. The following table lists the number of partitions of zero with a given area, up to an area of 50. A bijective toolkit for signed partitions 19 Area Pop 0 1 1 1 2 1 3 3 4 2 5 3 6 5 7 4 8 6 Area Pop 9 10 10 6 11 10 12 16 13 10 14 15 15 24 16 16 17 25 Area Pop 18 31 19 26 20 33 21 46 22 38 23 49 24 62 25 55 26 66 Area Pop 27 79 28 72 29 96 30 107 31 96 32 121 33 34 141 131 35 159 Area 36 Pop 185 37 173 38 199 39 40 237 216 41 266 42 43 296 272 44 334 Area 45 Pop 376 46 349 47 417 48 49 465 448 50 512 The grouping is deliberate: observe the behavior of the areas by threes. Areas of size 3k + 1 are deficient in number compared to their two neighbors; areas of size 3k appear more often, and areas of size 3k + 2 even more, roughly as often as areas of size 3k + 4, one grouping up. Why? The areas under each half of these convex paths are half-integers that are halves of squares and truncated squares; the generating function for the area under any such path is X z Q h,k≥0 1≤j≤h 1≤l≤k (1 − z h2 +2hk−k2 2 h2 2 (h−j)2 − 2 q h−k +jk q j )(1 l2 − z − 2 q −l ) When k = 0 we get the analogous expression for ordinary partitions. Curious whether this phenomenon held for those ordinary partitions, calculations were made, but the half-integer weights for such paths grow slowly and a pattern was difficult to discern in early data. However, a simpler approach yielded similar behavior; consider the “second integral” of a partition to be the progressive sum λ1 + (λ1 + λ2 ) + (λ1 + λ2 + λ3 ) + . . . , or λ1 · k + λ2 · (k − 1) + · · · + λk · 1. (”Groups of people arrive in bunches, most at first and then fewer later. You have to put up the lot of them at a local hotel until they have all arrived. How many person-nights are you going to be stuck for?”) The name “second integral” is chosen as the inverse of the much-studied first differences, second differences, etc. of a partition, which are the analogues of differentiation for the partition viewed as a discrete, nonincreasing function on the natural numbers. (The first integral is its weight.) The total area thus produced can range from n to (n2 + n)/2. The table below gives the number of times areas arise: 20 William J. Keith Area Pop 0 1 1 1 2 1 3 2 4 1 5 2 6 3 7 2 8 2 Area Pop 9 4 10 3 11 4 12 5 13 3 14 5 15 7 16 4 17 7 Area 18 Pop 8 19 6 20 8 21 11 22 7 23 9 24 13 25 9 26 11 Area 27 Pop 16 28 12 29 15 30 18 31 13 32 17 33 20 34 17 35 21 Area 36 Pop 24 37 19 38 24 39 30 40 22 41 28 42 34 43 26 44 34 Area 45 46 47 48 49 50 Pop 38 30 37 43 37 42 The growth is slower, but exhibits similar behavior; indeed, 3k, 3k+2, and 3k+4 are now all roughly equal. Again, one would like to know why, and if the behaviors described persist indefinitely, persist indefinitely with exceptions, or later yield to domination by the general increase. This entire line of questioning has rather departed from anything involving signed partitions. However, since it arose due to this investigation, it is presented here as an item of interest rooted in Andrews’ happy revisiting of Euler’s overlooked partitions. 3.4. Profiles and Complements One valuable feature of the standard Ferrers diagram is the profile of a partition, the outer edge of the squares occupied by dots in its Ferrers diagram. If the partition has fewer than N parts, all of size less than M , it fits inside an N × M rectangle. A lattice path from the lower left corner of that rectangle (the point (0, −N ) if we consider the standard Ferrers diagram as being in the fourth quadrant) to the upper right corner (the point (M, 0) likewise) is a set of steps north and east in this rectangle, and every such path describes a partition with a given profile. The statistics of the partitions thus described, especially the weight, form a set of useful statistics on lattice paths. The number of partitions in the N × M rectangle with weight j is the coefficient of q j in the q-binomial coefficient · N +M M ¸ = q (q; q)N +M , (q; q)N (q; q)M so these ´ objects are analogues of the usual binomial coefficients, i.e. ³ h named i because N +M N +M → as q → 1. M M q Conjugation of partitions is a bijection between lattice paths in the N × M rectangle and the M × N rectangle. It reverses the order of all steps, then exchanges east for north steps and vice versa. The weight of the path is preserved but not, we see, its dimensions. A bijective toolkit for signed partitions 21 One involution on lattice paths in the N × M rectangle is the complement, in which we simply reverse the order of steps, or equivalently, rotate the rectangle a half turn and take the other side of the path as our new partition. This preserves dimensions ibut not (usually) h N +M weight. Indeed, it shows that the coefficients of the polynomial are symmetric M q about q N M/2 , an analytic identity that can be rendered · ¸ · ¸ N +M N +M = qN M . M M q q −1 (That latter base seems to suggest that any proof featuring this identity is ripe for a new look involving signed partitions. We can combinatorially interpret the above as producing partitions in the N × M rectangle by subtracting from an N × M required positive block a partition into negative parts that themselves fit in such a rectangle.) On the other hand, consider our original Ferrers diagram. A signed partition, say (λ1 , . . . , λj , λ−1 , . . . , λ−k ), displayed in such a fashion has a profile constituting a lattice path from (−λ−k , −k) to (λ1 , j), necessarily starting and ending with two north steps, and passing through the origin via two east steps unless one or both of the sides are empty. By removing those steps, which corresponds to removing the largest “hook” of the positive and negative portions of the partition, we obtain a correspondence between lattice paths in the (N − 2)x(M − 2) box and several signed partitions with exactly M parts, at least some negative and some positive, for which λ1 + λ−k = N . The correspondence is not a bijection – the east steps surrounding the origin can be placed on any node (i.e., separate the parts before and after using the two steps), meaning that each such path maps to exactly N + M − 3 signed partitions of the type described in the N × M rectangle. Unfortunately, the weight changes under this map in a fashion very dependent on each path, so it does not make a convenient way to count signed partitions in the N × M box. Instead, visualize the following: suppose we have a signed partition into at most M total parts, with at least one part negative and λ1 + λ−k ≤ N (λ1 = 0 if no parts are positive). Diagram it with our original Ferrers diagram. Up-justify the positive parts so that the height difference from top to bottom is exactly M ; draw a vertical line from the origin if no parts are positive. We have the profile of a partition in the N × M box, with at most M −1 parts. The weight of our signed partition is exactly the weight of this partition, less a negative rectangle of size M λ−k . Thus, call p± (N, M ; n) the number of signed partitions in the N ×M box of weight n. We then have N M X n=−N M · p± (N, M ; n)q n = N +M M ¸ + q N X m=1 · q −M m N +M −1 M −1 ¸ . q The map is illustrated in figure 10. It is obvious that, if λ is an ordinary partition n, the weight of λ will be the same as its complement, say c(λ), in any N × M rectangle that λ fits in with N M = 2n. It is of interest to study what partitions might be called “self-complementary,” but as studied in the previous literature this characteristic is defined with reference to a given N × M 22 William J. Keith F IGURE 10. A signed partition mapped to an ordinary partition in the 21x13 box. h i +M box, a choice external to the partition itself. Stanley showed that NM , evaluated at q = −1, gives the number of self-complementary partitions in the N × M rectangle. (Note: if a partition is to be self-complementary in any rectangle at all, it will be so in a rectangle of height equal to the number of parts, and length equal to the largest part plus the smallest part, or by conjugation of length equal to λ1 and height equal to λk plus the number of times λ1 is repeated. Suppose we simply ask how many partitions of n have the property that k(λ1 + λk ) = 2n or the conjugate box described has area 2n, and call these partitions complementable. The first complementable partition that is not self-complementary in the associated box occurs for n = 15. It is far too tangential to this paper to discuss here, but might be another question for research.) The relevance of this discussion in this paper is that a simple variant of our basic complementation transformation always preserves the weight of a signed partition! Namely, let λ = (λ1 , . . . , λj , λ−1 , . . . , λ−k ) be a signed partition, and let N = max(λ1 , −(λ−k )), M = max(j, k). Then both the positive parts and the negative parts are partitions that fit in an N ×M or larger rectangle. Take the complement of the positive parts in the N × M rectangle, and reverse the sign of all parts to produce negative parts γ−1 , . . . , γ−m . Take the complement of the negative parts in the same size rectangle, and again reverse the sign of the resulting parts to create positive parts γ1 , . . . , γl . Gather these as a new signed partition γ. Suppose the weight of the positive parts of λ totals A, and the weight of the negative parts totals −B. Then |λ| = A−B, and |γ| = (−N M +A)+(N M −B) = |λ|. A signed partition is self-complementary under this map if the positive and negative portions are themselves complements of each other in some rectangle, which is naturally the rectangle to be chosen. Self-complementary partitions of 0 of height n are, then, given exactly by complementable partitions of n. 3.5. Issues for Exploration Any newly formalized field of inquiry has an embarrassment of riches when it comes to open questions. Some of the more interesting questions I have come across in preparing A bijective toolkit for signed partitions 23 F IGURE 11. A complementation in the 5 × 5 rectangle; the weight, n = −4, is preserved. this paper, which might be worth an investigator’s time, include the following tangents to questions discussed above: • When is a class of signed partitions a “natural” intermediary between two classes of ordinary partitions? • What combinatorially interesting objects are enumerated naturally by classes of signed partitions? A good example is conjugacy classes in SL(n, Z); the Jordan form of the matrices will have blocks of lengths partitioning n, each having 1 or -1 on the main diagonal; thus, these are described by signed partitions wherein the sum of the absolute value of parts is n. • How many complementable and self-complementary signed and ordinary partitions of n are there? Provide some nice generating functions for these and describe how the two sets differ. • Explain the behavior of partitions of zero in their arithmetic progressions modulo 3 and 6. • Explore the connection between lattice paths in the (N − 2)x(M − 2) rectangle, and the signed partitions in the N × M box to which they map under the possible placements of the origin. • Partitions of type B, with size restricted, are addressed by the first Borwein Conjecture. Could our map, or a similar one, make headway toward this open question? The author wishes to thank Dr. Andrews for reference to a number of papers which added both useful resources and several useful questions this paper could tackle. As the list of issues above shows, there is much more to be explored in nearby regions of this new way of looking at partitions. References [1] Alladi, K; Andrews, G.E.; and Gordon, B.: Generalizations and refinements of a partition theorem of Gollnitz. J. Reine Angew. Math. 460:165-188 (1995) [2] Andrews, G.E. Euler’s “De Partitio Numerorum.” Bull. of the Amer. Math. Soc. 44(4):561-573 (2007) 24 William J. Keith [3] Andrews, G.E. Schur’s theorem, partitions with odd parts and the Al-Salam-Carlitz polynomials. AMS Contemporary Mathematics, 254:45-53 (2000) [4] Andrews, G.E. Two theorems of Gauss and allied identities proved arithmetically. Pacific J. Math. 41:563-578 (1972) [5] Andrews, G.E. The Theory of Partitions, The Encyclopedia of Mathematics and Its Applications Series, Addison-Wesley Pub. Co., NY, 300 pp. (1976). Reissued, Cambridge University Press, New York, 1998. [6] Bessenrodt, C. A bijection for Lebesgue’s partition identity in the spirit of Sylvester. Discrete Math. 132 (13) 1-10 (1994) [7] Garrett, Kristina C.: A determinant identity that implies Rogers-Ramanujan. Electronic Journal of Combinatorics 12 (2005). [8] Keith, William J. Ranks of Partitions and Durfee Symbols. Ph.D. Thesis, Pennsylvania State University, 2007. Published online at http://etda.libraries.psu.edu/theses/approved/WorldWideIndex/ETD-2026/index.html [9] Pak, Igor. Partition bijections, a survey. The Ramanujan Journal 12 (1), August 2006, pp. 5-75. URL: http://dx.doi.org/10.1007/s11139-006-9576-1 [10] Stockhofe, D.: Bijektive Abbildungen auf der Menge der Partitionen einer naturlichen Zahl. Bayreuth. Math. Schr. (10), 1-59 (1982) [11] Yee, Ae Ja: A combinatorial proof of Andrews’ partition functions related to Schur’s partition theorem, Proceedings of the Amer. Math. Soc., 130 (2002), 2229-2235 [12] Želobenko, D.P. Compact Lie Groups and their Representations. Translations of Mathematical c Monographs, v. 40, °1973 American Mathematical Society. [13] Zeng, J.: The q-variations of Sylvesters bijection between odd and strict Partitions.” Ramanujan Journal 9(3), 289303 (2005) William J. Keith Drexel University 3141 Chestnut Philadelphia, PA 19104 USA e-mail: wjk26@drexel.edu