On products of long cycles: short cycle dependence and separation probabilities
aa r X i v : . [ m a t h . C O ] J un ON PRODUCTS OF LONG CYCLES:SHORT CYCLE DEPENDENCE AND SEPARATION PROBABILITIES
VALENTIN FÉRAY AND AMARPREET RATTANA
BSTRACT . We present various results on multiplying cycles in the symmetric group. One result is a gen-eralisation of the following theorem of Boccara (1980): the number of ways of writing an odd permutationin the symmetric group on n symbols as a product of an n -cycle and an ( n − -cycle is independent ofthe permutation chosen. We give a number of different approaches of our generalisation. One partial proofuses an inductive method which we also apply to other problems. In particular, we give a formula for thedistribution of the number of cycles over all products of cycles of fixed lengths. Another application is re-lated to the recent notion of separation probabilities for permutations introduced by Bernardi, Du, Moralesand Stanley (2014).
1. I
NTRODUCTION
We begin by introducing standard definitions and notation.Let n be a positive integer. A partition λ of n is a weakly decreasing sequence of positive integers ( λ , · · · , λ r ) whose sum is n , and we write λ ⊢ n or | λ | = n . We call the λ i the parts of λ . Thenumber of parts r is the length of λ and we write this as ℓ ( λ ) . We will also use the exponential notation λ = (1 m , m , . . . ) , where m i (or m i ( λ ) ) is the number of parts of λ equal to i . For any non-negativeinteger a < n , a partition of the form ( n − a, a ) is called a hook .Let S n be the symmetric group on the symbols [ n ] := { , , . . . , n } . It is well known that theconjugacy classes of S n are indexed by partitions of n ; a permutation’s weakly decreasing list of cyclelengths determines the partition giving its conjugacy class. We call this partition the permutation’s cycletype , and we denote by C λ the class of permutations with type λ . Any permutation whose cycle type is ahook is called a cycle , and, more specifically, a permutation with cycle type ( n − i, i ) is an ( n − i ) -cycle.A full cycle has type ( n ) .Let ε be the signature of a permutation; that is, ε ( σ ) = 1 if σ is an even permutation and − otherwise.We abuse notation and write ε ( λ ) for a partition λ , which is the signature of any permutation of type λ ,and we call the partition even or odd accordingly. It is well known that a partition λ is even if and onlyif | λ | − ℓ ( λ ) is an even integer.For a partition of λ of n , let K λ be the formal sum of all elements in the group algebra of S n withcycle type λ . The set { K λ : λ ⊢ n } forms a basis for the centre of the group algebra. Thus, for partitions µ and ν of n , we have an expansion K µ · K ν = X λ ⊢ n c λµ,ν K λ . The numbers c λµ,ν are known as the connection coefficients of S n and are the central objects of thisarticle. Equivalently, the connection coefficient c λµ,ν gives the number of ways of writing a fixed permu-tation of cycle type λ as the product of two permutations of cycle type µ and ν , respectively. It is thislatter combinatorial interpretation of the connection coefficients that we focus on in this article.Finding c λµ,ν is an important problem in algebraic combinatorics. See for example the early work ofH. Farahat and G. Higman [FH59], D. Walkup [Wal79], R. Stanley [Sta81], D. Jackson [Jac87] and I.Goulden and D. Jackson [GJ92]. The connection coefficients of the symmetric group also appear incontexts outside of algebra or combinatorics. In particular, P. Hanlon, J. Stembridge and R. Stanleyrelate them to the spectra of random matrices in [HSS92], while D. Jackson and T. Visentin [JV90]establish a connection with rooted maps embeddable on orientable surfaces. A rather general formula for c λµ,ν is due to A. Goupil and G. Schaeffer [GS98]. But this result, likemost results before it, requires that at least one of the three partitions λ, µ or ν is ( n ) . Since | C λ | c λµ,ν = | C µ | c µλ,ν , we see it is immaterial which one of λ, µ or ν is ( n ) .The reason why many of the known results require one of λ, µ or ν to be ( n ) is that a common tool tocompute these numbers, which is central in a number of the articles above, is the well-known expressionrelating the connection coefficients to the irreducible character values of the symmetric group, containedin Proposition 2.1 below. This expression, which is not specific to the symmetric group, is in generalnot tractable for arbitrary partitions. When one of the partitions is ( n ) , however, there is a substantialsimplification in the formula given in Proposition 2.1: only characters indexed by hooks contribute tothe result. This well-known simplification is contained in Lemma 2.2 below.The starting point of this paper is the following observation: when one of the partitions correspondsto an n -cycle and another corresponds to an ( n − a ) -cycle for small a (say µ and ν , respectively), theformula expressing c λµ,ν in terms of characters becomes even simpler.For partitions λ ⊢ n and ρ ⊢ a with n ≥ a , we denote by ν ρ ( λ ) := c λ ( n − a,ρ ) , ( n ) ; i.e. ν ρ ( λ ) is thenumber of ways to write a given permutation σ of type λ as the product of a n -cycle α and a permutation β of type ( n − a ) ∪ ρ . We present the following theorem. Theorem 1.1.
Let n and a < n be positive integers. Fix a partition ρ of a . There exists an explicitpolynomial Z ρ in the variables n and m , . . . , m a − such that, for every partition λ of n with m i cyclesof length i (for ≤ i ≤ a − ), one has that (1) ν ρ ( λ ) = (cid:0) − a ε ( ρ ) ε ( λ ) (cid:1) ( n − a − Z ρ ( n, m , . . . , m a − ) . That ν ρ ( λ ) vanishes if ( − a ε ( ρ ) ε ( λ ) = − is the consequence of a simple parity argument onpermutation signs. The surprising aspect of Theorem 1.1 is that ν ρ ( λ ) depends only on | λ | , ε ( λ ) , m ( λ ) ,. . . , m a − ( λ ) and not on the larger multiplicities of λ .Theorem 1.1 can be seen as a generalisation of Theorem 1.2 by Boccara [Boc80]. Theorem 1.2 (Boccara) . Let n be a positive integer and λ an odd partition of n . Then c λ ( n ) , ( n − , = 2( n − . A combinatorial proof of Theorem 1.2 is given in Cori, Marcus and Schaeffer [CMS12].Theorem 1.1 can be established using characters of the symmetric groups, which we do in Section 2.It is natural to look for a combinatorial explanation of it. We provide two approaches to this question: • in Section 3, we give a purely bijective proof of Theorem 1.1 in the case ρ = (1 a ) , extendingthe work of Cori et al. [CMS12]; • in Section 4, we present an inductive proof (assuming Boccara’s theorem), also for the case ρ = (1 a ) , using only combinatorial arguments.It seems that the case ρ = (1 a ) has a particular structure and we shall refer to it as the hook-case . Findinga combinatorial proof for a general partition ρ remains an open problem.Interestingly, each method allows to compute explicitly ν a ( λ ) for small values of a and leads todifferent expressions for ν a ( λ ) . For instance, our inductive approach leads to a compact expression of ν a ( λ ) in terms of symmetric functions; see Theorem 4.4. In addition, the case where λ has no parts oflength smaller than a , except fixed points, leads to a particularly compact and elegant formula (provedin Section 3.4). Proposition 1.3.
Let a ≥ be an integer and σ a permutation in S n . Assume σ has no cycles of lengthsmaller than a , except fixed points (whose number is denoted by b ). The number of ways to write σ asthe product of an n -cycle and an ( n − a ) -cycle is (cid:0) − a ε ( σ ) (cid:1) ( n − a − n − b )! a !( n − a − b + 1)! . N PRODUCTS OF CYCLES 3
An advantage of combinatorial proofs is that they can often be used to refine enumerative results,taking into account statistics that cannot be studied easily with the character approach. We illustrate thisfact by studying separation probabilities in products of n -cycles with ( n − a ) -cycles (here, we can alsodeal with the case a = 0 ).The notion of separable permutations has been introduced in a recent paper of Bernardi, Du, Moralesand Stanley [BDMS14]. The main question is to compute the probability that given elements are inthe same or different cycles in a product of uniform random permutations of given types (see Section 5for a formal definition). For example, in Proposition 5.7, we give an explicit formula for the followingproblem, formulated by Stanley in [Sta10, pages 54-71]: what is the probability that , , . . . , k lie indifferent cycles of α · β , where α and β are a random full cycle and an ( n − a ) -cycle, respectively, in S n ? Our method is to establish an induction relation for separation probabilities (Theorem 5.6). Thisformula is based on the inductive proof of the hook-case Theorem 1.1.Another advantage of the inductive method is that, in principle, it can be generalised to factorisa-tion problems where we do not require any partition to be ( n ) , although computations become morecumbersome. We give two examples of this kind of results in Section 6: • an explicit formula for the distribution of the number of cycles in the product of two cycles ofgiven lengths (Theorem 6.9); • an involved explicit formula for the separation probability for the product of two ( n − -cycles(Theorem 6.11). This leads us to an appealing conjecture (Conjecture 6.13), extending a sym-metry property proved by the four authors mentioned above, given in [BDMS14, Eq. (1)].Our final comment is that our methods are elementary and combinatorial. With the exception of Sec-tion 2 where we give our character explanation of Theorem 1.1, and Section 4.3 where we give a resultin the ring of symmetric functions, we do not use algebraic tools or combinatorial objects significantlydifferent from factorisations in the symmetric group. Outline of the paper.
Sections 2, 3 and 4 are devoted to the three approaches of Theorem 1.1: arepresentation-theoretic proof for general ρ , a bijective proof for the hook-case, and an inductive proofalso for the hook-case, respectively. In addition, at the end of Section 4, we give a formula for a sym-metric function related to multiplying cycles in the symmetric groups.In Section 5, we compute the separation probability of the product of a full cycle and a ( n − a ) -cycle.In Section 6.1, we present a number of lemmas needed for the inductive method. These are then usedto prove the theorems in Section 6 mentioned above.2. C HARACTER EXPLANATION
Character values and connection coefficients.
Let us consider the family of symmetric groups S n (for each n ≥ ). It is well known (see, e.g. , [Sag01, Chapter 2]) that both conjugacy classesand irreducible representations of S n can be indexed canonically by partitions of n , so the charactertable of S n is a collection of numbers χ λ ( µ ) , where λ and µ are taken over all partitions of n and are,respectively, the indices of the irreducible representation and the conjugacy class.While the following formula is difficult to attribute to a particular author, character values of sym-metric groups are a classical tool to compute connection coefficients of the symmetric group (see [JV90,Lemma 3.3]). Proposition 2.1.
For any triple of partitions ( λ, µ, ν ) of the same integer positive n , one has that (2) c λµ,ν = n ! z µ z ν X π ⊢ n χ π ( λ ) χ π ( µ ) χ π ( ν ) χ π (1 n ) , where z µ is the integer Q i ≥ i m i ( µ ) m i ( µ )! . This proposition has been widely used in the last thirty years to obtain explicit expressions for con-nection coefficients, especially in the case where one of the partitions λ , µ or ν is ( n ) ; see e.g. [GS98]and references therein. VALENTIN FÉRAY AND AMARPREET RATTAN
Lemmas on character values.
Two common methods of computing the characters of the symmet-ric group are found in the literature. One method is to use a change of basis in the symmetric functionring [Fro00], and the second is the Murnaghan-Nakayama combinatorial rule, which is in fact due toLittlewood and Richardson [LR34, § 11] (for a modern treatment see [Sag01] or [Mac95, § 1.7]).The following lemma is already well known.
Lemma 2.2.
Let π be a partition of n , one has: χ π (( n )) = 0 , unless π is a hook, that is ( n − i, i ) for some i between and n − . This is an immediate consequence of Murnaghan-Nakayama rule and its use can found in almost allof the aforementioned articles that use Proposition 2.1 to determine c λµ,ν . A consequence of Lemma 2.2is that by setting µ = ( n ) in Proposition 2.1, the sum over all partitions of n reduces to a sum overhooks, and thus a simple sum of i from to n − , which is usually much easier to handle.When ν = ( n − a ) ∪ ρ , where ρ is a partition of a fixed integer a , the sum also simplifies due to thefollowing lemma. Lemma 2.3.
Let i and n be two integers with ≤ i < n , and π be the hook partition ( n − i, i ) . Also,let a < n be a positive integer and ρ a partition of a . Assume further a − < i < n − a . Then, χ π (cid:0) ( n − a ) ∪ ρ (cid:1) = 0 . Proof.
The Murnaghan-Nakayama rule implies that χ π (cid:0) ( n − a ) ∪ ρ (cid:1) = 0 , unless we can find a ribbon ξ of size n − a in the diagram of π , given by n − i − i , so that π with ξ removed remains a Young diagram. As a > , the ribbon is not the whole diagram.Hence it is either in the first row (which is possible only if n − a ≤ n − i − ) or in the first column(which is possible only if n − a ≤ i ). If neither of these conditions is satisfied, there is no ribbon of size ( n − a ) in the diagram of π and the corresponding character value vanishes, as asserted. (cid:3) The last lemma we need is due to Littlewood [Lit50, p. 139]. See also [Sta81, Lemma 2.2]:
Lemma 2.4.
Let i and n be two integers with ≤ i < n , λ a partition of n , and π the hook partition ( n − i, i ) . Then χ π ( λ ) = X ρ ⊢ i ( − r + r + ... (cid:18) m − r (cid:19)(cid:18) m r (cid:19) . . . (cid:18) m i r i (cid:19) , where m j (respectively r j ) is the multiplicity of j in λ (respectively ρ ) for ≤ j ≤ i . Corollary 2.5.
With the notation of Lemma 2.4 above, if we fix the sign ε ( λ ) , then χ π ( λ ) is a polynomialin the multiplicities m j for j ≤ i, n − i − .Proof. The case i ≤ n − i − comes from the explicit expression above. The case i > n − i − followsfrom the symmetry formula, found in [Mac95, §1.7, Example 2], for characters χ π ( λ ) = ǫ ( λ ) χ π ′ ( λ ) , where π ′ = ( i + 1 , n − i − ) is the conjugate partition of π . (cid:3) We now have the necessary tools to give a proof of Theorem 1.1.
N PRODUCTS OF CYCLES 5
Proof of Theorem 1.1.
Fix a partition ρ of a positive integer a . By Proposition 2.1, for any partition λ of size n bigger than a , one has ν ρ ( λ ) = c λ ( n ) , ( n − a ) ∪ ρ = n ! n · ( n − a ) · z ρ X π ⊢ n χ π ( λ ) χ π (cid:0) ( n ) (cid:1) χ π (cid:0) ( n − a ) ∪ ρ (cid:1) χ π (1 n ) . By Lemmas 2.2 and 2.3, the only non-zero terms of the sum correspond to hook partition π = ( n − i, i ) ,where i is an integer which fulfils either i ≤ a − or i ≥ n − a .But, for such partitions π , the dimension χ π (1 n ) is (cid:0) n − i (cid:1) , and a simple use of the Murnaghan-Nakayama rule shows that χ π (( n )) = ( − i . Another application Murnaghan-Nakayama rule gives: • for i ≤ a − , one has that χ π (cid:0) ( n − a ) ∪ ρ (cid:1) = χ ˆ π ( ρ ) ; where ˆ π = ( a − i, i ) because the onlyribbon of size n − a is contained in the first row of π . • for i ≥ n − a , one has that χ π (cid:0) ( n − a ) ∪ ρ (cid:1) = ( − n − a − χ ˆ π ( ρ ) where ˆ π = ( n − i, i − n + a ) because the only ribbon of size n − a is contained in the first column of π .By Lemma 2.4, χ π ( λ ) depends on ε ( λ ) and in a polynomial way on m ( λ ) , . . . , m a − ( λ ) . For each i ≤ a − or i ≥ n − a , the quotient n ! n ( n − a ) (cid:0) n − i (cid:1) is ( n − a − multiplied by a polynomial in n , so this completes the proof of Theorem 1.1. (cid:3) Small values.
Using Lemma 2.4, we can also obtain explicit expressions. For example, using thefollowing notation for falling powers ( n ) m := n ( n − · · · ( n − m + 1) , for small values of a , we get: ν ( λ ) = (1 + ε ( λ ))( n − n − m ( λ )2 ; ν ( λ ) = (1 − ε ( λ ))( n − n + m ( λ ) −
22 ; ν ( λ ) = (1 − ε ( λ ))( n − (cid:0) ( n − − m − n −
2) + ( m − − m (cid:1) .
3. B
IJECTIVE APPROACH
Fix a permutation σ ∈ S n of type λ ⊢ n with the same parity as a . In this section, we will provebijectively Theorem 1.1 in the case ρ = 1 a ; that is, up to a factor ( n − a − , the number of ways towrite σ as a product α · β , where α is a n -cycle and β a ( n − a ) -cycle depends polynomially on n, m ( λ ) , · · · , m a − ( λ ) and not on the higher multiplicities of λ . The proof is a generalisation of the argument given in [CMS12]in the case a = 1 .Note that α = σ · β − is entirely determined by β (recall that σ is fixed), hence the question can bereformulated as follows: count the number of ( n − a ) -cycles β such that σ · β − is a full cycle. Example . Take the partition λ = (3 , , of n = 7 and a = 2 . We fix σ = (1 2 3)(4 5)(6 7) . Thenthe -cycle β = (1 3 7 5 2) fulfils the condition above. Indeed, α = σ · β − = (1 3 2 4 5 6 7) is a fullcycle.3.1. A necessary and sufficient condition. An ( n − a ) -cycle β can be written as follows(3) β = ( b )( b ) · · · ( b a − )( b a · · · b n − ) , where the b i are distinct integers between and n . While it may seem strange to write explicitly a − fixed points of β in equation (3) instead of all of them, this is central to our construction. Example . For β = (1 3 7 5 2) , we can choose b = 4 , b = 1 , b = 3 , b = 7 , b = 5 and b = 2 . This is of course not the only possible choice. VALENTIN FÉRAY AND AMARPREET RATTAN F IGURE
1. Graphs of Examples 3.3, 3.4, 3.10 and of the proof of Lemma 3.11.We would like α = σ · β − to be a n -cycle. To ensure this condition, we look at the graphicalrepresentation of α . By definition, it is the directed graph with vertex set [ n ] and directed edges ( i, α ( i )) for i ∈ [ n ] . This graph is always a union of cycles, corresponding to the cycles of α . Example . The graphical representation of σ = (1 2 3)(4 5)(6 7) is the left-most graph of Figure 1.The n − oriented edges(4) ( b i −→ σ ( b i ) for ≤ i ≤ a − b i −→ σ ( b i − ) for a + 1 ≤ i ≤ n − are some of the edges of the graphical representation of permutation α . Example . Choose b i as above (for i = 1 , . . . , ). Then the edges describedin (4) are drawn on the middle-left graph of Figure 1. They clearly form a subgraph of the graphicalrepresentation of α = (1 3 2 4 5 6 7) , which is the right-most graph in Figure 1 (for now, disregard thecolouring).The following lemma is an extension of [CMS12, Proposition 1]. Lemma 3.5.
Fix an integer a and a permutation σ of the same parity as a . Let b , . . . , b n − be distinctintegers between and n and let β be the permutation defined by equation (3) . Set also α = σβ − .Then α is a n -cycle if and only if the edges (4) form an acyclic set of edges.Proof. If α is a n -cycle, its graphical representation is a cycle and, thus, every strict subgraph is acyclic.It is in particular the case for the set (4) of edges.In the other direction, let us assume acyclicity of (4); that is, the graphical representation of α containsan acyclic subset of edges of size n − . Hence, it can be either a cycle or the union of two cycles. But, α has the same sign of a n -cycle ( β is an ( n − a ) -cycle, and the sign of σ corresponds to the parity of a )and hence the second possibility never occurs. (cid:3) Counting sequences.
Let us now try to enumerate sequences b , . . . , b n − as above.We first enumerate sequences b , . . . , b a − such that the edges b i → σ ( b i ) do not form any cycle.Such a sequence is equivalent to the ordered choice, in the graphical representation of σ , of a − edgesthat do not contain any cycle. Lemma 3.6.
Let G be a directed graph with n vertices and n edges, where G is a disjoint union of m i cycles of length i for i ≥ . Then the number N a ( σ ) of acyclic subsets of a − edges is given by (5) N a ( σ ) = a − X k =0 X n ,...,na − ≥ n ··· +( a − na − k ( − n + ··· + n a − (cid:18) n − ka − − k (cid:19) a − Y i =1 (cid:18) m i n i (cid:19) . In particular, when a is fixed, the above equation is a polynomial in n , m , m , . . . , m a − .Proof. Let us label the cycles of the graph by integers from to r and denote by λ j the length of cycle j . Fix a subset J of [ r ] . The number of sets of a − edges containing the cycle labelled j for all j in J is: (cid:18) n − k J a − − k J (cid:19) , where k J = X j ∈ J λ j . N PRODUCTS OF CYCLES 7
We use the convention that the binomial coefficient vanishes if k J > a − . By inclusion-exclusion, thenumber of sets of a − edges containing no cycles is: X J ⊆ [ r ] ( − | J | (cid:18) n − k J a − − k J (cid:19) . We split the sum depending of the value of k J and on the number n i of j in J such that λ j = i . Doingthis, we obtain (5) (note that if n i > for some i > a − , then k J > a − and the corresponding termvanishes). (cid:3) Finally, the number of possible values for the list ( b , . . . , b a − ) is ( a − N a ( σ ) .The enumeration of the possible values of b i , for i ≥ a , is harder. A method due to Lehman is toenumerate such sequences with an additional element x (which is also an integer between and n ), withthe condition that the edge b a → σ ( x ) does not create a cycle when added to the graph in (4). Definition . A Lehman sequence of type a for σ is a sequence ( x, b , . . . , b n − ) of integers between and n such that: • the b i ( ≤ i ≤ n − ) are distinct (but x can be equal to some of the b i ); • the following set of n − edges is acyclic:(6) b i −→ σ ( b i ) for ≤ i ≤ a − b i −→ σ ( b i − ) for a < i ≤ n − b a −→ σ ( x ) . Example . With σ and b i (for i = 1 , · · · , ) defined as in previous ex-amples, we can choose any x ∈ { σ − (3) , σ − (2) , σ − (4) , σ − (5) , σ − (6) } so that ( x, b , · · · , b ) is aLehman sequence of type for σ . Indeed, the edge (1 , σ ( x )) must not create a cycle when added to themiddle-left graph of Figure 1.In general, we shall see later that the number of possible values for x (when b , . . . , b n − are fixed) isthe oriented distance from b a to f in the graphical representation of α , where f is the integer between and n different from all the b i (in our example, b a = 1 and f = 6 ).With the extra parameter x , Lehman sequences are easy to enumerate. The following is an extensionof [CMS12, Proposition 2]). Lemma 3.9.
The number of Lehman sequences of type a for σ is n · ( a − · N a ( σ ) · ( n − a )! . As a corollary, for a fixed a ≥ , it depends polynomially on n , m ,. . . , m a − and not on higher multi-plicities.Proof. We first choose an ( a − -tuple ( b , · · · , b a − ) such that the edges b i → σ ( b i ) do not contain acycle. We saw that there are ( a − · N a ( σ ) choices for this list. Then, choose any number between and n as the value for x .Let us compute the number of possible values of b a , after we have fixed the values of x, b , . . . , b a − .By definition of a Lehman sequence, b a must be different from all values b , . . . , b a − . We must alsoavoid forming a cycle with the edge b a → σ ( x ) in the graph with edges b i → σ ( b i ) . This forbids oneadditional value, the end of the (possibly empty) maximal path beginning at σ ( x ) in this graph (thismaximal path is unique as at most one edge leaves every vertex). Finally there are n − a possible valuesfor b a .The same proof shows that, if b a , · · · , b j − are fixed, there are n − j possible values for b j (for all j ∈ { a + 1 , · · · , n − } ). This ends the proof of the lemma. (cid:3) Example . [Illustration of the proof] Take as before σ = (1 2 3)(4 5)(6 7) , b = 4 , b = 1 and b = 3 . We also set x = 5 , that is σ ( x ) = 4 . As a consequence we know that (6) contains themiddle-right graph of Figure 1.We want to find the number of possible values b = b , b , b such that the edge b → σ ( b ) = 1 does not add any cycle to the configuration. In addition to the forbidden values { , , } , one should VALENTIN FÉRAY AND AMARPREET RATTAN also avoid b = 5 , which would create a cycle (1 , , . So the number of possible values for b is , asasserted in the proof above.3.3. From sequences to permutations.
Because of Lemma 3.5, to each Lehman sequence we canassociate a permutation β of type ( n − a, a ) , such that σ · β − is a full cycle. For fixed n and σ , denotethe function mapping a Lehman sequence of type a for σ by Λ a . We now prove that Λ a is k -to- (anextension of [CMS12, Proposition 3]), for some appropriate number k . Lemma 3.11.
Let β be a permutation of type ( n − a, a ) , such that σ · β − is a full cycle. Set k = ( a − · a ( n − a )2 · n Then there are k Lehman sequences of type a that are pre-images of β by Λ a .Proof. Fix a permutation β as in the statement of the lemma. A Lehman sequence ( x, b , . . . , b n − ) oftype a for β is given by the following: • the choice of which fixed point of β does not appear in the list b , . . . , b a − (we will denote it f ); • the choice of an element b a in the ( n − a ) -cycle of β (the values of b a +1 , b a +2 , . . . follow fromthis choice); • the choice of x ; • a choice of a permutation of the fixed points b , . . . , b a − .For the last item, the number of possible permutations is ( a − and does not depend on the choicesabove. However, the number of choices for x does depend on f and b a , so we enumerate directly thenumber of possible triplets ( x, f, b a ) .We consider the graphical representation of α = σ · β − . By assumption, this is a directed cycle oflength n . We can colour its vertices as follows (denote G α ; β the resulting coloured graph): • the fixed points of β are vertices coloured in blue (for readers of black-and-white printed ver-sions, these are squared vertices). • the vertices corresponding to points in the support of β are vertices coloured in red ( circular vertices).For example, if β = (1 3 7 5 2) and α = (1 3 2 4 5 6 7) , then G α ; β is the right-most graph of Figure 1.With this colouring, f and b a must be respectively chosen among the blue and red vertices of G α ; β .The graph with edges (4) is obtained from G α ; β by erasing the edges leaving the blue vertex f and thered vertex b a . This graph is a disjoint union of two paths, one ending in b a and one ending in f (seeExample 3.4 where b a = 1 and f = 6 ).Now, x must be chosen such that b a → σ ( x ) does not create a cycle, that is σ ( x ) must be on the pathending in f . Thus, if b a and f are fixed, there are d ( b a , f ) possible values for x , where d is the oriented distance from b a to f in G α ; β (see Example 3.8).Finally the number of triplets ( x, f, b a ) that yield a Lehman sequence is(7) X vb ∈ Vbvr ∈ Vr d ( v r , v b ) , where V b and V r are respectively the set of blue and red vertices of G α . Note that d ( v r , v b ) − is thenumber of vertices v (of any colour), which are on the path between v r and v b (and different from v r and v b ). So expression (7) can be rewritten as | V b | · | V r | + (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) ( v b , v r , v ) , s.t. v b is blue ; v r is red ; v is on the path from v r to v b . (cid:12)(cid:12)(cid:12)(cid:12)(cid:12)(cid:12) But any set of three distinct vertices of G α ; β with either two blue and one red vertices or two red andone blue vertices can be seen in a unique way as a triplet of the set above. So finally (7) is equal to: | V b | · | V r | + (cid:18) | V b | (cid:19) · | V r | + | V b | · (cid:18) | V r | (cid:19) = | V b | · | V r | | V b | + | V r | ) . As | V b | = a , while | V r | = n − a , this ends the proof of the lemma. (cid:3) N PRODUCTS OF CYCLES 9
Finally, Lemmas 3.9 and 3.11 imply the following. If ε ( λ ) = ( − a , then the number of ( n − a ) -cycles β such that σβ − is a cycle is(8) ν a ( λ ) = n · ( a − · N a ( σ ) · ( n − a )!( a − · a ( n − a )2 · n = 2 ( n − a − a N a ( σ ) . The case ρ = (1 a ) of Theorem 1.1 follows from Lemma 3.6.3.4. Small values.
Combining Eqs. (5) and (8), the proof above gives explicit formulae for ν (1 a ) ( λ ) ,slightly different from the one obtained with the character approach. For instance, ν ( λ ) = (1 + ε ( λ )) ( n − n − m ( λ )); ν ( λ ) = (1 − ε ( λ )) ( n − (cid:18)(cid:18) n (cid:19) − ( n − · m + (cid:18) m (cid:19) − m (cid:19) . The expressions look slightly different from the one obtained in Section 2.4, but they are of courseequivalent, as it can be easily checked by the reader.We can now give a quick proof of Proposition 1.3, which follows directly from the above bijectiveproof. If σ has no cycles of length smaller than a , except b fixed points, then it is clear that N a ( σ ) = (cid:0) n − ba − (cid:1) . This yields the formula claimed in the introduction.4. D ETERMINING CONNECTION COEFFICIENTS USING INDUCTION RELATIONS
Another proof of Theorem 1.1.
We now give an induction proof of Theorem 1.1. For any integerpartition λ , let λ ↓ ( j ) be the partition obtained from λ by removing a part of size j and replacing it witha part of size j − . We do not use this notation if λ has no part of size j . Lemma 4.1.
Let λ ⊢ n + 1 and a < n a non-negative integer. Then ( a + 1) | C λ | ν a +1 ( λ ) = ( n + 1) X j ≥ mj ( λ ) > | C λ ↓ ( j ) | ν a (cid:16) λ ↓ ( j ) (cid:17) · ( m j − ( λ ) + 1) · ( j − , (9) and ( a + 1) ν a +1 ( λ ) = X j ≥ mj ( λ ) > ν a (cid:16) λ ↓ ( j ) (cid:17) · m j ( λ ) · j. While we will establish in Section 6 a more general result (Corollary 6.6), we elect to prove Lemma 4.1directly immediately below to make the first four sections self-contained.
Proof.
Inspecting the left hand side of (9), we see that it counts triples ( σ, α, β ) in S n +1 with thefollowing properties: σ · α · β = e , where e is the identity element; σ , α , and β are of types ( n + 1) , λ ,and ( n − a, a +1 ) , respectively; and β has a distinguished fixed point. Notice that α and β cannot sharea fixed point because σ does not have any, and their product is e . The right hand side of (9) counts theunion of quadruples ( s, ¯ σ, ¯ α, ¯ β ) over { j ≥ m j ( λ ) > } with the following properties: s ∈ [ n + 1] ; ¯ σ, ¯ α and ¯ β are permutations of [ n + 1] \ { s } of types ( n ) , λ ↓ ( j ) and ( n − a, a ) , respectively with ¯ σ · ¯ α · ¯ β = e ; and ¯ α has a member of a ( j − -cycle distinguished. We give a bijection between thesetwo sets.Suppose that ( σ, α, β ) is a triple counted by the left hand side (9), and suppose that s is the distin-guished fixed point of β . Denote by j the length of the cycle of α containing s . Obviously, m j ( λ ) > and, besides, j ≥ as s cannot be a fixed point of α . Simply remove s from σ , α and β to obtainpermutations ¯ σ, ¯ α and ¯ β , respectively, and record the image of s in α . It’s not difficult to see that ¯ σ , ¯ α and ¯ β have the claimed cycle type. Removing s from those permutations is equivalent to defining ¯ σ , ¯ α and ¯ β as the restrictions of ( s σ ( s )) σ , ( s α ( s )) α and ¯ β , respectively, to [ n + 1] \ { s } . Furthermore,we easily compute ¯ σ · ¯ α · ¯ β = ( s σ ( s )) · σ · ( s α ( s )) · α · β = ( s σ ( s )) · ( σ ( s ) σ ◦ α ( s )) · σ · α · β = e, where the third inequality follows from the fact that σ ◦ α ( s ) = s . The construction is clearly reversible, establishing the first assertion of the lemma. For the second, itis well known that | C λ | = ( n + 1)! Q i i m i ( λ ) m i ( λ )! , whence we obtain, for j ≥ with m j ( λ ) > , ( n + 1) | C λ ↓ ( j ) || C λ | = j · m j ( λ )( j − m j − ( λ ) + 1) . Therefore (9) becomes ( a + 1) ν a +1 ( λ ) = X j ≥ mj ( λ ) > ν a ( λ ↓ ( j ) ) · m j ( λ ) · j, completing the proof of the second assertion of the lemma. (cid:3) We now prove Theorem 1.1 in the case ρ = 1 a . Proof of Theorem 1.1 for ρ = 1 a . Our proof proceeds by induction on a . By Theorem 1.2 (Boccara’sTheorem), we have ν ( λ ) = (1 − ε ( λ ))( n − , establishing the base case.Now suppose a > . By induction, there exists a polynomial G a − ( x , . . . , x a − ) such that for anyparticular partition λ with | λ | > a , we have ν a − ( λ ) = (cid:0) − a − ε ( λ ) (cid:1) ( | λ | − a )!( a − G a − ( | λ | , m ( λ ) , . . . , m a − ( λ )) . For a fixed λ , we use Lemma 4.1 to obtain that ν a ( λ ) = 1 a X j ≥ mj ( λ ) > ν a − ( λ ↓ ( j ) ) · m j ( λ ) · j = 1 a X j ≥ mj ( λ ) > (1+( − a − ε ( λ ↓ ( j ) ))( | λ ↓ ( j ) |− a )!( a − G a − ( | λ ↓ ( j ) | , m ( λ ↓ ( j ) ) , . . . , m a − ( λ ↓ ( j ) )) · m j ( λ ) · j. (10)For j ≥ , we see that ε ( λ ↓ ( j ) ) = − ε ( λ ) and | λ ↓ ( j ) | = | λ | − . Thus, we see (10) becomes (1+( − a ε ( λ ))( | λ |− a − a ! a − X j ≥ mj ( λ ) > G a − ( | λ | − , . . . , m j − ( λ ) + 1 , m j ( λ ) − , . . . , m a − ( λ )) · m j ( λ ) · j + X j ≥ amj ( λ ) > G a − ( | λ | − , m ( λ ) , . . . , m a − ( λ )) · m j ( λ ) · j = (1+( − a ε ( λ ))( | λ |− a − a ! a − X j ≥ mj ( λ ) > G a − ( | λ | − , . . . , m j − ( λ ) + 1 , m j ( λ ) − , . . . , m a − ( λ )) · m j ( λ ) · j + G a − ( | λ | − , m ( λ ) , . . . , m a − ( λ )) | λ | − a − X j =1 mj ( λ ) > m j ( λ ) · j . Notice that in the two equations immediately above that the condition m j ( λ ) > is superfluous becausethere are no longer terms with λ ↓ ( j ) , and the terms with m j ( λ ) are 0. The last equation contains a N PRODUCTS OF CYCLES 11 polynomial dependent on | λ | , and m i ( λ ) for ≤ i ≤ a − , but otherwise independent of λ , completingthe proof. (cid:3) Small values.
Notice that the above proof gives an inductive formula to compute ν a ( λ ) . For λ ⊢ n , we have ν ( λ ) = (1 + ε ( λ ))( n − n − m ( λ )); ν ( λ ) = (1 − ε ( λ ))( n − n − m ( λ )) − m ( λ )); ν ( λ ) = (1 + ε ( λ ))( n − n − m ( λ )) − m ( λ )( n − m ( λ ) − − m ( λ )) . Once again, the expressions look different from the ones obtained by the other method (Sections 2.4 and3.4), but they are of course equivalent.4.3.
A symmetric function formula.
We end this section with a symmetric function formula. Thisformula will not be used in this paper, but we mention it because it encodes our induction in a very com-pact way. As this is not central in the paper, we do not recall needed definitions that involve symmetricfunctions. We will, however, use the standard notation found in [Mac95, Chapter I].Let Λ be the symmetric function ring. It admits a linear basis, called power-sum basis ( p λ ) λ , indexedby all partitions. Let C [ S n ] be the group algebra of S n . Let us consider the linear operator ψ : C [ S n ] → Λ σ n ! p λ , where λ is the cycle-type of σ . This is a natural transformation, which is used in particular to linkirreducible characters of the symmetric group with Schur functions (see [Mac95, section I,7]). Recallfrom Section 1 that K λ is the sum of permutations of type λ and lies in the centre of C [ S n ] .The purpose of this section is to give an explicit expression for F a ( n ) := ψ (cid:0) K ( n ) · K ( n − a, a ) (cid:1) interms of monomial symmetric functions ( M λ ) .From the definitions, we have that F a ( n ) := ψ (cid:0) K ( n ) · K ( n − a, a ) (cid:1) = 1 n ! X λ ⊢ n | C λ | ν a ( λ ) p λ . As always in this paper, the case a = 1 is easy. For a = 1 , one has ν ( λ ) = (1 − ε ( λ ))( n − .Denoting as usual z λ = n ! / | C λ | , one has(11) F ( n ) = ( n − X λ ⊢ n (1 − ε ( λ )) p λ z λ = ( n − X λ ⊢ n p λ z λ − X λ ⊢ n ε ( λ ) p λ z λ ! = ( n − X λ ⊢ nλ =(1 n ) M λ , where the last equality comes from [Mac95, Chapter I, Eq (2.14’)].We then proceed by induction using to the following lemma. Lemma 4.2.
For a ≥ , F a +1 ( n + 1) = 1 a + 1 ∆ (cid:0) F a ( n ) (cid:1) , where ∆ is the differential operator X i x i ∂∂x i . Proof.
Fix a ≥ . On the one hand, one has that F a +1 ( n + 1) = 1( n + 1)! X µ ⊢ n +1 | C µ | ν a +1 ( µ ) p µ = n + 1( n + 1)! · ( a + 1) X µ ⊢ n +1 p µ X j ≥ mj ( µ ) > (cid:12)(cid:12)(cid:12) C µ ↓ ( j ) (cid:12)(cid:12)(cid:12) ν a (cid:16) µ ↓ ( j ) (cid:17) · ( m j − ( µ ) + 1) · ( j − , where the second equality comes from Lemma 4.1.Let λ ↑ ( j ) the partition obtained from λ by replacing a part of size j with a part of size j + 1 (again,the notation is defined only if m j ( λ ) > ). The function ( µ, j ) ( µ ↓ ( j ) , j ) is clearly bijective from theset { ( µ, j ) , µ ⊢ n + 1 , j ≥ , m j ( µ ) > } to { ( λ, j ) , λ ⊢ n, j ≥ , m j − ( λ ) > } – the inverse is ( λ, j ) ( λ ↑ ( j − , j ) – so we can change the summation indices. F a +1 ( n + 1) = 1 n ! · ( a + 1) X λ ⊢ n | C λ | ν a ( λ ) X j ≥ mj − λ ) > (cid:0) m j − ( λ ↑ ( j − ) + 1 (cid:1) · ( j − · p λ ↑ ( j − = 1 n ! · ( a + 1) X λ ⊢ n | C λ | ν a ( λ ) X j ≥ mj − λ ) > m j − ( λ ) · ( j − · p λ ↑ ( j − . But, on the other hand, it is easy to check that X j ≥ mj − λ ) > m j − ( λ ) · ( j − · p λ ↑ ( j − = ∆( p λ ) . A detailed proof of the previous equation can be found in [FV12, Section 6.1]; the reader uncomfortablewith infinitely many variables should also look at this reference, where this issue is discussed. Finally,using the linearity of the operator ∆ , we get that F a +1 ( n + 1) = 1 a + 1 ∆ n ! X λ ⊢ n | C λ | ν a ( λ ) p λ ! , finishing the proof. (cid:3) We need one more lemma before proceeding to the main theorem of this section.
Lemma 4.3.
The operator ∆ is injective.Proof. As ∆ is homogeneous (it sends symmetric functions of degree n to symmetric functions of degree n + 1 ), it is enough to prove the lemma for the restriction to symmetric functions of a specific degree n .Consider a linear combination P λ ⊢ n a λ M λ sent by ∆ onto . Using the description of the action of ∆ on the monomial basis given in [FV12, Section 6.1], we get X λ ⊢ n a λ X j> mj ( λ ) > j · ( m j +1 ( λ ) + 1) · M λ ↑ ( j ) = 0 . Rewriting this as usual as a sum over µ and j , we get X µ ⊢ n +1 M µ X j> mj +1( µ ) > a µ ↓ ( j +1) · j · m j +1 ( µ ) = 0 . This equation is equivalent to the following linear system: for any µ ⊢ n + 1 ,( E µ ) X j> mj +1( µ ) > a µ ↓ ( j +1) · j · m j +1 ( µ ) = 0 Fix λ ⊢ n and µ = ( λ + 1 , λ , λ , . . . ) . Then equation ( E µ ) involves the variable a λ and variables a ν with ν ⊢ n and ν = λ + 1 . In particular, such partitions ν are lexicographically bigger than λ .In other terms, the linear system ( E µ ) µ ⊢ n +1 admits a triangular subsystem and the only solution is a λ = 0 for all λ ⊢ n . (cid:3) N PRODUCTS OF CYCLES 13
We can now obtain an explicit formula for F a ( n ) . By convention, we set ( x ) − = x +1 , which iscompatible with the relation ( n ) m · ( n − m ) = ( n ) m +1 . Theorem 4.4.
For any n > and a ≥ , one has that F a ( n ) = ( n − a − a ! X λ ⊢ n | λ |− ℓ ( λ ) ≥ a ( | λ | − ℓ ( λ )) a − M λ . Proof.
Let us first look at the case a ≥ . We proceed by induction. For a = 1 , the theorem correspondsto equation (11). Let us suppose that the theorem is true for a fixed a ≥ . Then, by Lemma 4.2 andinduction hypothesis, one has: F a +1 ( n + 1) = 1 a + 1 ∆ (cid:0) F a ( n ) (cid:1) = ( n − a − a + 1)! X λ ⊢ n | λ |− ℓ ( λ ) ≥ a ( | λ | − ℓ ( λ )) a − ∆( M λ ) . But, as explained in [FV12, Section 6.1], ∆( M λ ) = X j> mj ( λ ) > j · ( m j +1 ( λ ) + 1) · M λ ↑ ( j ) . Substituting this into the equation above, we have that F a +1 ( n + 1) = ( n − a − a + 1)! X λ ⊢ n | λ |− ℓ ( λ ) ≥ a X j> mj ( λ ) > ( | λ | − ℓ ( λ )) a − · j · ( m j +1 ( λ ) + 1) · M λ ↑ ( j ) . We use the same manipulation as in the proof of Lemma 4.2: the double sum on λ ⊢ n and j can beturned into a double sum on µ ⊢ n + 1 and j where µ = λ ↑ ( j ) . Note that the size and length of µ arerelated to λ by | µ | = | λ | + 1 and ℓ ( µ ) = ℓ ( λ ) . We get that F a +1 ( n + 1) = ( n − a − a + 1)! X µ ⊢ n +1 | µ |− ℓ ( µ ) ≥ a +1 X j> ( | µ | − ℓ ( µ ) − a − · j · ( m j +1 ( µ )) · M µ = ( n − a − a + 1)! X µ ⊢ n +1 | µ |− ℓ ( µ ) ≥ a +1 M µ · ( | µ | − ℓ ( µ ) − a − X j> j · ( m j +1 ( µ )) = ( n − a − a + 1)! X µ ⊢ n +1 | µ |− ℓ ( µ ) ≥ a +1 M µ · ( | µ | − ℓ ( µ )) a . This ends the proof of the case a ≥ .For a = 0 , the same computation as above shows that ∆ ( n − X λ ⊢ n | λ | − ℓ ( λ ) + 1 M λ ! = ( n − X µ ⊢ n +1 | µ |− ℓ ( µ ) ≥ M µ = F ( n + 1) . But, by Lemma 4.2, one also has ∆( F ( n )) = F ( n +1) . Hence the theorem follows from the injectivityof ∆ , given in Lemma 4.3. (cid:3) Remark . The case a = 0 of Theorem 4.4 was already known, under a slightly different form; see[Vas12, Theorem 2]. Remark . Theorem 1 of [Vas12] also contains a symmetric function involving all connection coef-ficients c ( n ) µ,λ ; We can show our symmetric function formula contained in Theorem 4.4 can be obtainedfrom [Vas12, Theorem 1] via involved λ -ring manipulations. The presentation above, however, is justas short and doesn’t rely on the non-trivial result [Vas12, Theorem 1].
5. S
EPARATION P ROBABILITIES
Definition . Let J be a set partition ( J , . . . , J k ) of [ m ] of length k . A permutation σ ∈ S n (with n ≥ m ) is called J -separated if no cycle of σ contains two elements from different blocksof J . It is said to be strongly J -separated if, moreover, each J i is contained in some cycle of σ .It is easy to see that highly symmetric sets ( i.e. sets of permutations whose sum lie in the centre ofthe group algebra of S n ), that the number of J -separated permutations only depends on the composition I = ( i , . . . , i k ) of m , where i p = J p . By convention, a permutation is I -separated for a composition I , if it is ( J , . . . , J k ) -separated, where J p = { i p − + 1 , . . . , i p } . We use the two notions of separationgiven by a set partition and separation given by a composition interchangeably.The same convention holds for strongly separable permutations.As mentioned at the end of the paper [BDMS14], the problem of computing separation probabilitiesor strong separation probabilities are linked to each other by a simple relation (equation (34) of the citedpaper). Therefore, we shall use the notion depending on which one is most convenient. Note also, thatfor I = (1 k ) (the case for which we have explicit expressions), the two notions coincide.The rest of Section 5 is devoted to computing the strong separation probability of the product ofa random n -cycle with a random ( n − a ) -cycle, whereas in Section 6.3 we use the standard notionof separable permutations. Therefore when we talk about separable permutation in the remainder ofSection 5, we will mean strongly separable.An induction formula for these quantities is given in Section 5.4. To do that, we need some prelim-inary results regarding the computation of separation probabilities in simpler models: uniform randompermutations (Section 5.1), uniform random odd permutations (Section 5.2) and uniform random per-mutations of a given type (Section 5.3).5.1. Uniform random permutation.
Let I = ( i , . . . , i k ) be a composition of m of length k . Thefollowing result can be found in [Sta10, p. 13], but we copy it here for completeness.The probability P I that a uniform random permutation in S n ( n ≥ m ) is strongly I -separated isgiven by(12) P I = ( i − · ( i − · · · ( i k − m ! . Proof.
The result is obvious for n = m . Indeed, if J is a set-partition of [ m ] , a permutation in S m is J -separated if it is the disjoint product of a cycle of support J ( ( i − choices), a cycle of support J ( ( i − choices) and so on.We then proceed by induction on n . A random permutation of n + 1 can be obtained from a randompermutation σ of n by choosing uniformly an integer j between and n + 1 and • either add n + 1 as a fixed point if j = n + 1 ; • or add n + 1 right after j in the cycle notation of σ .Both operations do not change the fact that the permutation is J -separated, so the probability of being J separated does not depend on n . (cid:3) Uniform random odd permutation.
Let I be a composition of m of length k . The probabilities P In, odd and P In, even for a uniform random odd (respectively even) permutation in S n ( n > m ) to be I -separated fulfils the relation:(13) P In, odd = n − n ( P In − , even ) + 1 n P In − , odd . Proof.
Let
Odd n and Even n be the set of odd and even permutation in S n , respectively. Then, one hasa bijection φ : Odd n ≃ Odd n − ∪ ([ n − × Even n − ) , where φ and its inverse are given by the following rules. • For σ ∈ S n , set φ ( σ ) to be the restriction of σ to [ n − if n is a fixed point of σ ; otherwise,let j = σ − ( n ) = n and set φ ( σ ) = ( j, τ ) , where τ is obtained by erasing n in the cycledecomposition of σ . N PRODUCTS OF CYCLES 15 • φ − ( τ ) is its canonical embedding in S n if τ is an odd permutation of [ n − ; otherwise φ − ( j, τ ) = τ ( j n ) for ( j, τ ) ∈ [ n − × Even n − .Clearly, in both cases, σ is I -separated if and only if τ is. A quick computation leads to the formulaabove. (cid:3) As a random permutation in S n has probability P I to be I -separated, one has: P In, odd + P In, even = 2 P I . Thus (13) can be made into the induction relation: P In, odd = n − n (2 P I − P In − , odd ) + 1 n P In − , odd . To solve this induction, let us define Q In = ( P In, odd − P I ) . Our induction then becomes Q In = − n − n Q In − Together with the base case (the argument to compute P Im, odd is the same than in the proof of (12)) Q Im = ( − P I if m − k is even, P I − P I if m − k is odd,we obtain that Q In = ( − n − k +1 P I m ( m − n ( n − . Therefore, we have proved the following proposition (this result is also given in [Sta10, p. 61]).
Proposition 5.2.
Let I be a composition of m of length k and n an integer with n ≥ m . The probabilityfor a random odd permutation in S n to be I -separated is P In, odd = P I · (cid:18) − n − k +1 m ( m − n ( n − (cid:19) . Uniform random permutation of a given type λ .Proposition 5.3. Let I = ( i , . . . , i k ) be a composition of m of length k and λ a partition of n of length r . The probability that a random permutation of type λ is I -separated is given by (14) P I ( λ ) = 1( n ) m X ≤ j ,...,jk ≤ r distinct k Y t =1 ( λ j t ) i t . Proof.
Each partition π of type λ can be written in exactly z λ := n ! / | C λ | ways as π = (cid:0) b . . . b λ (cid:1)(cid:0) b λ +1 . . . b λ + λ (cid:1) . . . , with b , . . . , b n containing exactly once each number from to n . We denote C j ( λ ) the set of indices inthe j -th cycle of π , that is C j ( λ ) := (cid:8) λ + · · · + λ j − + 1 , . . . , λ + · · · + λ j (cid:9) . Fix some distinct integers j , . . . , j k between and r . As in the introduction of the section, we denote J h = { i h − + 1 , . . . , i h } . Let us consider the property: “for each h , the numbers in J h lie in the j h -thcycle of π ”.For this property to be true, each b − ( J h ) must be included in C h ( λ ) , which means that there are ( λ j h ) i h possible values for the list ( b − ( i h − + 1) , . . . , b − ( i h )) . Thus, in total, there are Q kt =1 ( λ j t ) i t possible values for ( b − (1) , . . . , b − ( m )) such that the considered property holds. As there are in total ( n ) m possible values for this list and as this list is uniformly distributed when we take π uniformly atrandom in C λ , the probability that our property holds is n ) m k Y t =1 ( λ j t ) i t . For different sequences j , . . . , j k these events are incompatible and their union correspond to the factthat π is strongly J -separated, whence the result follows. (cid:3) Let us denote R I ( λ ) = ( n ) m P I ( λ ) the sum in equation (14) (it is a polynomial function in λ , λ , . . . ).The following lemma will be useful in the next section. Lemma 5.4.
Let I = ( i , . . . , i k ) be a composition of m of length k . For any partition λ of n of length r , one has that r X t =1 λ t R I ( λ + δ t ) = ( n + m ) R I ( λ ) + k X h =1 i h ( i h − R I − δ h ( λ ) , where δ t is the vector with only components, except for its t -th component which equals .Proof. Let us fix t ∈ [ r ] and compare R I ( λ + δ t ) and R I ( λ ) . The distinct summation indices j , . . . , j k obtained from (14) in R I ( λ + δ t ) and R I ( λ ) are the same and the summands are different only in thecase where one of the j s , let us say j h , is equal to t . Thus, one has that R I ( λ + δ t ) − R I ( λ ) = k X h =1 X j ,..., b j h ,...,j k = t (cid:0) ( λ t + 1) i h − ( λ t ) i h (cid:1) Y g = h ( λ j g ) i g . The hat in the summation index means that we sum over choices of integers j g for g = h . But, bydefinition: X j ,..., b j h ,...,j k = t Y g = h ( λ j g ) i g = R I \ i h ( λ \ λ t ) . Thus, noticing that ( λ t + 1) i h − ( λ t ) i h = i h ( λ t ) ( i h − , we obtain: R I ( λ + δ t ) − R I ( λ ) = k X h =1 i h ( λ t ) ( i h − R I \ i h ( λ \ λ t ) . Now we multiply by λ t and sum this equality over t : r X t =1 (cid:0) λ t R I ( λ + δ t ) (cid:1) − n · R I ( λ ) = r X t =1 k X h =1 i h λ t ( λ t ) ( i h − R I \ i h ( λ \ λ t )= k X h =1 i h r X t =1 (cid:0) ( λ t ) i h + ( i h − λ t ) ( i h − (cid:1) R I \ i h ( λ \ λ t ) ! . It is clear from the definition of R I that these functions satisfy r X t =1 ( λ t ) i R I ( λ \ λ t ) = R I ∪ i ( λ ) . Indeed, we simply split the sum in R I ∪ i ( λ ) depending on which part of λ is associated to the part i of I ∪ i . Returning to our computation, one has that r X t =1 (cid:0) λ t R I ( λ + δ t ) (cid:1) − n · R I ( λ ) = k X h =1 (cid:0) i h R I ( λ ) + i h ( i h − R I \ i h ∪ ( i h − ( λ ) (cid:1) , completing the proof. (cid:3) Product of a uniform random n -cycle and a uniform random ( n − a ) -cycle. Define π Iλ,µ tobe the probability that the product of two uniformly selected random permutations of types λ and µ ,respectively, is strongly I -separated. For ease of notation, let P In,a := π I ( n ) , ( n − a, a ) .The case a = 1 is particularly simple. Indeed, using Theorem 1.2, this is equivalent to choosing aodd permutation uniformly at random. Therefore, the computation of Section 5.2 gives that(15) P In, = P In, odd = P I · (cid:18) − n − k +1 m ( m − n ( n − (cid:19) . We will then establish an induction relation for P In,a . Let us begin with an elementary lemma.
N PRODUCTS OF CYCLES 17
Lemma 5.5.
Let I be a composition of an integer m and n and a be non-negative integers with n > a and n ≥ m . P In,a = a !( n − a )( n − n ! X λ ⊢ n | C λ | ν a ( λ ) P I ( λ ) . Proof.
The probability that the product α · β , where α and β are an n -cycle and ( n − a ) -cycle respectively,has cycle type λ is given by | C λ | ν a ( λ )( n − n ! a !( n − a ) Of course, all permutations of a given type λ have the same probability to occur. Therefore, the condi-tional probability that αβ is I -separated knowing that it has cycle type λ is P I ( λ ) . Putting everythingtogether, we obtain the formula above. (cid:3) Theorem 5.6.
Let I = ( i , . . . , i k ) be a composition of an integer m and n and a be non-negativeintegers with n ≥ m . One has that n ( n + 1) P In +1 ,a +1 = ( n − m + 1)( n + m ) P In,a + k X h =1 i h ( i h − P I − δ h n,a . Proof.
By lemma 5.5, one has that P In +1 ,a +1 = ( a + 1)!( n − a ) n !( n + 1)! X µ ⊢ n +1 | C µ | ν a +1 ( µ ) P I ( µ ) . We now use Lemma 4.1 and obtain: P In +1 ,a +1 = ( a + 1)!( n − a ) n !( n + 1)! X µ ⊢ n +1 n + 1 a + 1 X j ≥ mj ( µ ) > | C µ ↓ ( j ) | ν a (cid:0) µ ↓ ( j ) (cid:1) ( j − m j − ( µ ) + 1) P I ( µ ) We use the same reasoning as in the proof of Lemma 4.2: the function ( µ, j ) ( µ ↓ ( j ) , j ) is bijective –the inverse is ( λ, j ) ( λ ↑ ( j − , j ) – so we can change the summation indices. P In +1 ,a +1 = a !( n − a ) n ! n ! X λ ⊢ n X j ≥ mj − λ ) > ( j − m j − ( λ ) | C λ | ν a ( λ ) P I ( λ ↑ ( j − )= a !( n − a ) n ! n ! X λ ⊢ n | C λ | ν a ( λ ) ℓ ( λ ) X t =1 λ t P I ( λ + δ t ) . The last inequality comes from the fact that, for a fixed partition λ if we consider the multiset formed by ( j − m j − ( λ ) copies of λ ↑ ( j − for each j with m j − ( λ ) > , it is the same as the multiset with λ t copies of λ + δ t for each t between and ℓ ( λ ) .Now, notice that we have already computed in Lemma 5.4 the sum over t appearing in the previousequation (recall that P I ( λ ) = R I ( λ ) / ( n ) m ). We obtain: P In +1 ,a +1 = a !( n − a ) n ! n ! X λ ⊢ n | C λ | ν a ( λ ) · ( n + m )( n − m + 1) n + 1 P I ( λ ) + k X h =1 i h ( i h − n + 1 P I − δ h ( λ ) ! . We can now use Lemma 5.5 again and we get that P In +1 ,a +1 = ( n + m )( n − m + 1) n ( n + 1) P In,a + 1 n ( n + 1) k X h =1 i h ( i h − P I − δ h n,a ! . (cid:3) Recall that P In, is given by the simple formula (15). Using this, Theorem 5.6 allows us to compute P In,a for a ≥ by induction on a and P In, by induction on | I | − ℓ ( I ) . We end this section by giving afew applications of this fact. The induction on a for a ≥ becomes particularly simple in the case I = (1 k ) . In this case, m = k and the sum in Theorem 5.6 vanishes. We therefore find that n ( n + 1) P (1 k ) n +1 ,a +1 = ( n − k + 1)( n + k ) P (1 k ) n,a , from which we deduce immediately the following formula. Proposition 5.7.
Let n , a and k be positive integers with a, k ≤ n . Then P (1 k ) n,a = ( n − k ) a − · ( n − k ) a − ( n ) a − · ( n − a ) a − (cid:18) − n − k k ( k − n ( n + 1) (cid:19) . But, Theorem 5.6 is also valid for a = 0 and hence(16) ( n − m + 1)( n + m ) P In, = n ( n + 1) P In +1 , − k X h =1 i h ( i h − P I − δ h n, . Again, the case I = (1 k ) is particularly easy as the sum vanishes: we get that ( n − k + 1)( n + k ) P (1 k ) n, = n ( n + 1) 1 k ! (cid:18) − n − k k ( k − n ( n + 1) (cid:19) , which simplifies to(17) P (1 k ) n, = ( k ! if n − k is odd ; k ! (cid:16) k ( k − n − k +1)( n + k ) (cid:17) if n − k is even . This formula was established by R. Du and R. Stanley [Sta10, page 46], see also [BDMS14, Equation(2)]. For other compositions I , we are able to get information of the denominator of P In, , seen as afunction of n . Proposition 5.8.
Fix a composition I of size m and length k . Then, there exist polynomials Q eI ( n ) and Q oI ( n ) such that, for n ≥ m , P In, = ( Q eI ( n )( n − m +1)( n − m +2) ... ( n − k )( n − k +1)( n + k )( n + k +1) ... ( n + m ) if n − k is even Q oI ( n )( n − m +1)( n − m +2) ... ( n − k )( n + k +1) ... ( n + m ) if n − k is odd . Proof.
We proceed by induction over m − k .If m = k , that is if I = (1 k ) , the result follows from the explicit expression (17).For m > k , the result follows from Eq. (16), the induction hypothesis and the fact that n ( n +1) P In +1 , coincide with a polynomial in n for even (respectively odd) values of n . (cid:3) Some examples computed by Du and Stanley [Sta10, page 52] indicate that the above fractions arenot in their reduced form and that there are some nice cancellations that we did not manage to explainwith our induction formula.In another direction, after rescaling by defining(18) g P In,a = P In,a Q kh =1 ( i h − , separation probabilities exhibit the same kind of dependency that we have observed in Theorem 1.1. Asfor partitions, we denote m i ( I ) the number of parts i in I . Proposition 5.9.
Fix n and a ≥ . The quantity g P In,a depends on I only through its size m , its length k and its small multiplicities m ( I ) , . . . , m a − ( I ) .Proof (by induction on a ). The base case a = 1 follows from the explicit formula g P In,a = 1 m ! (cid:18) − n − k +1 m ( m − n ( n − (cid:19) , obtained from Eqs. (12), (15) and (18). N PRODUCTS OF CYCLES 19
Suppose that the statement holds for some a ≥ ; that is, the rescaled separation probability g P Jn,a isa function F n,a ( m ′ , k ′ , m ( J ) , . . . , m a − ( J )) , where m ′ and k ′ are respectively the size and length of J . Fix some composition I = ( i , . . . , i k ) with size m and length k . Then, Theorem 5.6, rewritten for g P In,a , yields: n ( n + 1) ^ P In +1 ,a +1 = ( n − m + 1)( n + m ) g P In,a + X ≤ h ≤ kih =1 i h ^ P I − δ h n,a . The first summand depends on I only through m , k , m ( I ) , . . . , m a − ( I ) by induction hypothesis, sowe focus on the last sum. We split it according to the value of i h : namely, set Σ b = X ≤ h ≤ kih = b i h ^ P I − δ h n,a = b m b ( I ) g P J b n,a . The last equality comes from the fact that, for a fixed b , all compositions J := I − δ h are reordering ofsome composition J b . Using the induction hypothesis, we get that g P J b n,a = F n,a (cid:0) m − , k, m ( I ) , . . . , m b − ( I ) + 1 , m b ( I ) − , . . . , m a − ( I ) (cid:1) for ≤ b ≤ a − F n,a (cid:0) m − , k, m ( I ) , . . . , m a − ( I ) , m a − ( I ) − (cid:1) for b = aF n,a (cid:0) m − , k, m ( I ) , . . . , m a − ( I ) (cid:1) for b > a. Therefore, for a fixed b ≤ a , the quantity Σ b depends on I only through m , k , m ( I ) , . . . , m a ( I ) .Moreover, we have X b>a Σ b = X b>a b m b ( I ) ! F n,a (cid:0) m − , k, m ( I ) , . . . , m a − ( I ) (cid:1) = m − X b ≤ a b m b ( I ) F n,a (cid:0) m − , k, m ( I ) , . . . , m a − ( I ) (cid:1) , so that this quantity also depends only on I only through m , k , m ( I ) , . . . , m a ( I ) . Hence the propositionholds for a + 1 . (cid:3) Remark . Weak separation probability (in opposition to strong separation probability as consideredhere) fulfils a much stronger symmetry property; see [BDMS14, Equation (1) ].6. P
RODUCTS OF CYCLES OF ARBITRARY LENGTH
Induction relations.
In this section, we present lemmas which will be needed in our inductiveapproach for product of cycles of arbitrary length. Our approach in this section will be similar to thatgiven in the proof of Lemma 4.1, but be more involved because it will apply more generally.Let λ and µ be partitions of n . Define A ( m, λ, µ ) to be the tuples ( σ, α, β ) in S n such that theirproduct satisfies σ · α · β = e , where e is the identity permutation, the permutations α, β have cycle type λ and µ , respectively, and σ has m cycles. Define also A i ( m, λ, µ ) to be the tuples in A ( m, λ, µ ) where α and β have precisely i common fixed points. In general, we use lower case a to mean the cardinality ofthe corresponding set with capital A (i.e. a i ( m, λ, µ ) = A i ( m, λ, µ ) and a ( m, λ, µ ) = A ( m, λ, µ ) , etc . ).It will be convenient in our induction to consider permutations of a set different from [ n ] . Denote S S the set of bijections from S to S . The notation above is naturally extended to the case of permutationswith ground set S by adding a superscript; that is, A S ( m, λ, µ ) (respectively A Si ( m, λ, µ ) ) is the set oftriples ( σ, α, β ) of permutations of the ground set S with the above properties.Before we begin our induction results, we make two simple observations regarding the sets A i ( m, λ, µ ) .For a partition λ with at least r λ by removing r
1s by λ [ r ] and, recall from Section 4, the partition λ ↓ ( j ) is obtained from λ by replacing a part of size j with a partof size j − if m j ( λ ) > . Thus, the two notations agree at λ ↓ (1) and λ [1] . Lemma 6.1.
Suppose that n ≥ and that ≤ r ≤ min { m ( λ ) , m ( µ ) } . Then A [ n ] r ( m, λ, µ ) = [ S ⊆ [ n ] | S | = r A [ n ] \ S ( m − r, λ [ r ] , µ [ r ] ) and a r ( m, λ, µ ) = (cid:18) nr (cid:19) a ( m − r, λ [ r ] , µ [ r ] ) . Proof.
The claim in the lemma is merely that the r common fixed points of α and β will also be fixedpoints of σ . We may therefore remove these common fixed points from the triple and consider the triple ( σ, α, β ) in A [ n ] \ S ( m − r, λ [ r ] , µ [ r ] ) , where S is the set of common fixed points. The second equationfollows from the first. (cid:3) Lemma 6.2.
For n ≥ , we have A ( m, λ, µ ) = min { m ( λ ) ,m ( µ ) } [ r =0 A r ( m, λ, µ ) Proof.
Follows from the definitions of the relevant objects. (cid:3)
We therefore see from Lemmas 6.1 and 6.2 that evaluating a ( m − r, λ [ r ] , µ [ r ] ) will help us obtainour formulae.Our main induction theorem will determine A ( m, λ, µ ) . We will find it useful to use rooted versionsof the factorisations introduced. Let A i ( m, λ, ˙ µ ) be the set of all triples ( σ, α, β ) ∈ A i ( m, λ, µ ) wherewe have distinguished one fixed point of β . Similarly, for j ≥ with m j ( λ ) > , let A i ( m, ˆ λ ↓ ( j ) , µ ) be the set of triples ( σ, α, β ) ∈ A ( m, λ ↓ ( j ) , µ ) , with one member of an ( j − -cycle distinguished, andwith precisely i fixed points in common. When j = 2 , this means that we distinguish a fixed point of α . If it is also a fixed point of β , we do not take it into account when counting the common fixed pointsof α and β (thus, we should say that α and β have precisely i non-rooted fixed points in common). Thefollowing decomposition holds. Lemma 6.3.
Suppose that µ, λ ⊢ n . If m ( µ ) = 0 , we have A [ n ]0 ( m, λ, ˙ µ ) = [ s ∈ [ n ] [ j ≥ mj ( µ ) > A [ n ] \ s ( m, ˆ λ ↓ ( j ) , µ [1] ) . Proof.
Define a function ψ mapping the left hand side to the right hand side by the following rule.Suppose that s is the rooted fixed point of β and suppose that s is contained in a j -cycle of α , for some j ≥ with m j ( λ ) > . Since α and β have no common fixed points, we know that α ( s ) = s . Define ¯ α as ( s α ( s )) α and ¯ β as β with the fixed point s removed. Also, let ¯ σ = ( s σ ( s )) σ . Finally, let ψ ( σ, α, β ) = (¯ σ, ¯ α, ¯ β ) , where the root of ¯ α is chosen to be α ( s ) . We show that ψ is well-defined andbijective.To show that ψ is well defined we must show the following. As defined, ¯ σ, ¯ α and ¯ β are permutationswith ground set [ n ] ; note, however, that s is clearly a fixed point of all of them, and thus we can considerthose permutations in S [ n ] \{ s } . Therefore, we must show that ¯ σ · ¯ α · ¯ β = e in S [ n ] \{ s } , ¯ α and ¯ β havecycle types λ ↓ ( j ) and µ [1] , respectively, and ¯ σ has m cycles. We show these in turn.The following computation show that the product of ¯ σ, ¯ α and ¯ β is the identity. ¯ σ · ¯ α · ¯ β = ( s σ ( s )) · σ · ( s α ( s )) · α · β = ( s σ ( s )) · ( σ ( s ) σ ◦ α ( s )) · σ · α · β = σ · α · β = e, where the third equality follows from s = σ ◦ α ◦ β ( s ) = σ ◦ α ( s ) . Thus, in S [ n ] we see that the productof ¯ σ, ¯ α and ¯ β is e , hence also true in S [ n ] \{ s } since they each fix s . N PRODUCTS OF CYCLES 21
Clearly, ¯ β is of the claimed cycle type, and we can additionally see that ( s α ( s )) removes s from thecycle in α that contains it and leaves the rest of the cycle unchanged, showing that ¯ α ∈ λ ↓ ( j ) for some j ≥ and m j ( λ ) > . To show that ¯ σ has m cycles, note that ( s σ ( s )) removes s from the cycle in ¯ σ that contains it, leaving the rest of the cycle unchanged. We should take care to ensure that s is notthe only member of its cycle in σ ; however, since α ( s ) = s and β ( s ) = s , we see that σ ( s ) = s , forotherwise the product of σ, α and β would not be the identity. Thus, we see that ¯ σ has m cycles. Thiscompletes the proof. (cid:3) We wish to apply the operations in Lemma 6.3 twice in succession. To do this we introduce thefollowing set. For i, j ≥ with m i ( λ ) , m j ( µ ) > , let A ( m, ˆ λ ↓ ( i ) , ˆ µ ↓ ( j ) ) be the set of triples ( σ, α, β ) ∈ A ( m, λ ↓ ( i ) , µ ↓ ( j ) ) such that a member of an i − and j − -cycle are roots of α and β ,respectively, and α and β have no non-rooted fixed points in common (thus, if i = 2 , we are to root afixed point of α and a rooted fixed point of α can occur as a fixed point in β ). We have the followinglemma. Lemma 6.4.
Let λ, µ ⊢ n and both m ( λ ) , m ( µ ) = 0 . Then, A [ n ]0 ( m, ˙ λ, ˙ µ ) = [ s,t ∈ [ n ] s = t [ i,j ≥ mi ( λ ) ,mj ( µ ) > A [ n ] \{ s,t } ( m, ˆ λ [1] ↓ ( i ) , ˆ µ [1] ↓ ( j ) ) . Proof.
Define a function ψ mapping the left hand side to the right hand side by the following rule.Suppose that s and t are the rooted fixed points of α and β , respectively.Define ¯ α = ( t α ( t )) α and ¯ β = ( s β ( s )) β . Also let p = ¯ α ◦ β ( s ) and define(19) ¯ σ = σ · ( t α ( t )) · ( s p ) . Finally, let ψ ( σ, α, β ) = (¯ σ, ¯ α, ¯ β ) . We choose as the root of ¯ α to be α ( t ) and for ¯ β we choose β ( s ) .To show that ψ is well defined we must show the following. As defined, ¯ σ, ¯ α and ¯ β are permutationswith ground set [ n ] ; we must, however, show that s, t are fixed points of all of them, and thus we canconsider those permutations in S [ n ] \{ s,t } . We must further show that ¯ σ ¯ α ¯ β = e in S [ n ] \{ s,t } , ¯ α and ¯ β have cycle types λ ↓ ( i ) and µ ↓ ( j ) , respectively, and ¯ σ has m cycles. The proof, however, largely followsthe proof of Lemma 6.3, so we omit the details. (cid:3) Note that it is straightforward to use Lemma 6.3 when m ( λ ) = 0 . In that case, it immediatelyfollows from Lemma 6.3 that if λ, µ ⊢ n with m ( µ ) = 0 , we have(20) m ( µ ) a ( m, λ, µ ) = n X j ≥ mj ( λ ) > a ( m, λ ↓ ( j ) , µ ↓ (1) ) · ( m j − ( λ ) + 1) · ( j − . If m ( λ ) = 0 , the complication is that a part of size 1 is created in λ ↓ (2) . This corresponds to a fixedpoint being created in a relevant permutation α and this fixed point may occur in β .Similarly, we can use easily Lemma 6.4 when m ( λ ) = m ( µ ) = 0 . If, furthermore, one has m ( µ ) , m ( λ ) = 0 , then(21) m ( λ ) m ( µ ) a ( m, λ, µ ) = ( n ) X i,j ≥ mi ( λ ) ,mj ( µ ) > a ( m, λ ↓ ( i )[1] , µ ↓ ( j )[1] ) · ( m i − ( λ ) + 1) · ( i − · ( m j − ( µ ) + 1) · ( j − . Note that this equation can be used only when m ( λ ) = m ( µ ) = 0 . We shall be mindful of thistechnicality, but will be able to overcome it easily.The reader has certainly noticed that the arguments used in this Section and in the proof of Eq. (9)are similar. In fact, the following variant of Lemma 6.3 and its Corollary generalizes Eq. (9). We will,however, primarily be using Lemmas 6.3 and 6.4 in the coming sections. Lemma 6.5.
Suppose that µ, λ ⊢ n , and m ( µ ) = 0 . Then A [ n ] ( m, λ, ˙ µ ) = [ s ∈ [ n ] [ j ≥ mj ( λ ) > A [ n ] \ s ( m, ˆ λ ↓ ( j ) , µ [1] ) ∪ [ s ∈ [ n ] A [ n ] \ s ( m − , λ [1] , µ [1] ) , where the last union should be omitted if m ( λ ) = 0 or m = 1 .Proof. The proof is similar to that of Lemma 6.3, except that a triple ( σ, α, β ) can all have the samefixed point. However, that is taken care of in the second union in the lemma. (cid:3) Counting the sets on both sides of Lemma 6.5 gives the following.
Corollary 6.6.
Suppose that λ, µ ⊢ n , and that m ( µ ) = 0 . Then, m ( µ ) a ( m, λ, µ ) = n X j ≥ mj ( λ ) > a ( m, λ ↓ ( j ) , µ ↓ (1) ) · ( m j − ( λ )+1) · ( j − n a ( m − , λ ↓ (1) , µ ↓ (1) ) , where the last term should be omitted if m ( λ ) = 0 or m = 1 . Eq. (9) is a special case of this Corollary 6.6 (with m = 1 and µ = ( n − a, a +1 ) ). We do not useCorollary 6.6 in its full generality in this paper. However, we hope that it can be used in the future togive generalisations of Lemma 4.1.6.2. A formula for multiplying cycles.
We use the ideas developed in Section 6.1, to obtain enumera-tive results for multiplying cycles; specifically, we look for formulae for a (cid:0) m, ( i + t, j − t ) , ( i, j )) (cid:1) for i ≥ and ≤ t ≤ j . To simplify notation, we will omit the parenthesis and commas in the notation ofthe partitions above: that is we shorten the notation to a ( m, i + t j − t , i j ) .First, note that, because of the sign function on S n , a ( m, i j , i + t j − t ) = 0 unless m + i + j + t is even.Second, the case j = t = 0 is known; see e.g. [Sta11, Corollary 3.4]. For any r ≥ and m , we have(22) a ( m, r, r )( r − ( c ( r +1 ,m ) ( r +12 ) if r − m is even otherwise,where c ( r + 1 , m ) is a signless Stirling number of the first kind; that is, it is the number of permutationson [ r + 1] with m cycles.Thus, we shall attempt to find a formula for a ( m, i + t j − t , i j ) in terms of a ( s, r, r ) for r ≤ i + j and s ≤ m . We first remark that we can easily determine a formula for i = 1 and ≤ t ≤ j , since a ( m, t j − t , j +1 ) = ( j + 1)!(1 + t )( j − t )! δ m,j − t +1 , where δ is the usual Kronecker δ function. We may therefore assume that i ≥ . We have the followingresults when j − t = 0 . Lemma 6.7.
Suppose that i ≥ and j ≥ . Then a ( m, i + j, i j ) = ( i + j ) j ( i + j − j j ! a ( m, i, i ) . Proof.
We prove the claim by induction. The claim is trivial when j = 0 , establishing the base case.Notice that a permutation of type ( i + j ) has no fixed points, so a ( m, i + j, i j ) = a ( m, i + j, i j ) .Now assuming that j > , note that i + j ≥ , and we can apply Lemma 6.3 as we did in (20). We getthat a ( m, i + j, i j ) = ( i + j ) · ( i + j − j a ( m, i + j − , i j − ) . Applying induction gives the desired formula. (cid:3)
Notice that if i − j + t < , or equivalently j > i + t , that a ( m, i + t j − t , i j ) = 0 . Keeping thatin mind, we have the following lemma. N PRODUCTS OF CYCLES 23
Lemma 6.8.
Suppose that i ≥ and ≤ t ≤ j ≤ i + t . If i − j + t ≥ , we have a ( m, i + t j − t , i j ) = ( i + j )!( i − j + t )! (cid:18) i − j − t (cid:19)(cid:18) i + t − j (cid:19) a ( m, i − j + t, i − j + t ) , and if i − j + t = 0 , we have a ( m, i + t j − t , i j ) = ( i + j )! i · ( i + t ) δ m, . Proof.
Substituting λ = ( r, n − r ) and µ = ( s, n − s ) in (21) yields that(23) a ( m, r n − r , s n − s ) = n ( n − · ( r − s − n − r )( n − s ) a ( m, r − n − − r , s − n − − s ) . But this equation is only valid if s, r ≥ . The idea is to iterate it as long as possible, and deal with thetwo special cases at the end.If i − j + t ≥ , we can iterate (23) without any complication. If a ( m, i j , i + t j − t ) is non-zero,we have(24) a ( m, i + t j − t , i j ) = ( i + j )( i + j − i − i + t − j ( j − t ) a ( m, i + t − j − t − , i − j − )= ( i + j )( i + j − i − i + t − j ( j − t ) · · · ( i − j + t + 2)( i − j + t + 1)( i − j + t )( i − j + 2 t )( t + 1)1 · a ( m, i − j + 2 t, i − j + t t )= ( i + j ) j − t ) ( i − j − t ( i + t − j − t ( j ) j − t ( j − t ) j − t a ( m, i − j + 2 t, i − j + t t ) . Therefore, we have a ( m, i + t j − t , i j ) = ( i + j ) j − t ) ( i − j − t ( i + t − j − t ( j ) j − t ( j − t ) j − t a ( m, i − j + 2 t, i − j + t t )= ( i + j )!( i − j + 2 t )! (cid:18) i − j − t (cid:19) ( i + t − i − j + 2 t − j ! t ! ( i − j + 2 t ) t ( i − j + 2 t − t t ! a ( m, i − j + t, i − j + t )= ( i + j )!( i − j + t )! (cid:18) i − j − t (cid:19)(cid:18) i + t − j (cid:19) a ( m, i − j + t, i − j + t ) . If i − j + t ≤ , we cannot use equation (21) iteratively as long as needed. We can, however, explicitlycount the final stages of the iteration given by a ( m, r n − r , s ) . In fact, only the cases r = n − and r = n − will be needed below.If r = n − , we have by explicitly counting a ( m, n − , n − ) = n !2( n − δ m, and if r = n − , we have a ( m, n − , n − ) = n ! δ m, = n ! a ( m, , . Thus, when i − j + t = 1 , we have by repeating the computation in (24) a ( m, i + t j − t , i j ) = ( i + j ) i − ( i − i − ( i + t − i − ( j ) i − ( j − t ) i − · a ( m, t + 2 1 , j − i +2 , )= ( i + j ) i − ( i − i − ( i + t − i − ( j ) i − ( j − t ) i − ( j − i + 4)! a ( m, , i + j )! a ( m, , , which is the claimed formula when i − j + t = 1 . If i − j + t = 0 , repeating the computation in (24)gives a ( m, i + t j − t , i j ) = ( i + j ) i − ( i − i − ( i + t − i − ( j ) i − ( j − t ) i − · a ( m, t + 2 1 , j − i +2 )= ( i + j ) i − ( i − i − ( i + t − i − ( j ) i − ( j − t ) i − ( j − i + 4)!2( j − i + 2) δ m, = ( i + j )! i · ( i + t ) δ m, . This completes the proof. (cid:3)
This lemma gives an explicit formula for a ( m, i j , i + t j − t ) . Theorem 6.9.
Suppose that i ≥ and ≤ t ≤ j , and further suppose that i + j + t + m is even. Then, a ( m, i + t j − t , i j ) = j − t X s =max { ,j − i − t +1 } ( i + j )! s !( i − j + t + s ) (cid:18) i − j − s − t (cid:19)(cid:18) i + t − j − s (cid:19) c ( i − j + t + s + 1 , m − s ) (cid:0) i − j + t + s +12 (cid:1) + Λ( m, i, j, t ) , where Λ( m, i, j, t ) = ( i − j + t ≥ ( i + j )!( j − i − t )! i · ( i + t ) δ m − j + i + t, if i − j + t ≤ . Proof.
We have from Lemmas 6.1 and 6.2, that(25) a ( m, i + t j − t , i j ) = X s (cid:18) i + js (cid:19) a ( m − s, i + t j − t − s , i j − s ) . Notice that if i − j + t ≤ , then a ( m − s, i + t j − t − s , i j − s ) = 0 unless s ≥ j − i − t . Hence therange for s in the theorem. Applying Lemma 6.8 to the summand in (25) and using formula (22) givesthe desired formula. (cid:3) We mention a particularly interesting case of multiplying two i -cycles in S n . Corollary 6.10.
Suppose that i ≥ , j ≥ and i + j + m is even. Then, a ( m, i j , i j ) = j X s =max { ,j − i +1 } ( i + j )! s !( i − j + s ) (cid:18) i − j − s (cid:19) c ( i − j + s + 1 , m − s ) (cid:0) i − j + s +12 (cid:1) + Λ( m, i, j ) , where Λ( m, i, j ) = ( i − j > ( i + j )!( j − i )! i δ m − j + i, if i − j ≤ . Separation probabilities when neither cycle is full.
In this section, we use the standard notionof separable permutations, as opposed to strong separation used in most of Section 5. In this section,we use Lemmas 6.1 and 6.4 to determine the separation probability for an arbitrary composition I whenmultiplying two ( n − -cycles in S n . To that end, let λ, µ ⊢ n and I = ( i , i , . . . i k ) be a compositionof m ≤ n . We use the notation of [BDMS14]: σ Iλ,µ is the probability that the product of two uniformlydistributed permutations of cycle types λ and µ , respectively, is I -separated. When λ = µ = ( n ) , thefollowing was proved in [BDMS14]:(26) σ I ( n ) , ( n ) = ( n − m )! Q ( i j )!( n − D m,kn , N PRODUCTS OF CYCLES 25 where D m,kn = ( n − n + k ) (cid:0) n + mm − k (cid:1) ( − n − m (cid:18) n − k − (cid:19) + m − k X r =0 (cid:18) n + mn + k + r (cid:19)(cid:18) n + r + 1 m (cid:19)! . We are chiefly interested in the case λ = µ = ( n ) or ( n − , . Let S Iλ be the number of permutationsthat are I -separated over all products of two permutations of cycle type λ . Thus, S I ( n ) = ( n − σ I ( n ) , ( n ) = ( n − m )! Y j ( i j )! D m,kn . Let A [ n ] ( λ ) = ∪ m A [ n ] ( m, λ, λ ) ; that is, A [ n ] ( λ ) it is set of all triples ( σ, α, β ) such that σαβ = e in S n and α and β have cycle type λ . Our goal is to determine the number of such triples in A [ n ] (( n − , where σ is I -separated. Using the decompositions in Lemmas 6.1 and 6.4 we seethat the elements in A [ n ] (( n − , come in two forms; either the triple ( σ, α, β ) has a common fixedpoint or it does not. We take these two cases separately, with the second case being a lot more involved.In speaking of separation below, we take J = ( J , . . . , J k ) to be the set partition of { , , . . . , m } ,where J p = { i p − + 1 , . . . , i p } . However, we remind ourselves that any set partition J whose blocksizes realise the composition I will have S Jλ = S Iλ . We use this fact repeatedly below.If ( σ, α, β ) has a common fixed point c , then by Lemma 6.1 these triples are in bijection with triples (ˆ σ, ˆ α, ˆ β ) in A [ n ] \ c (( n − . Thus, if c ∈ [ m ] , σ is J -separated if and only if ˆ σ is J \ c -separated, andif c / ∈ [ m ] then σ is J -separated if and only if ˆ σ is J -separated. We have a total of P kj =1 i j S I ↓ ( i j )( n − in theformer case, and(27) ( n − m ) S I ( n − in the latter case. The former simplifies as k X j =1 i j S I ↓ ( i j )( n − = k X j =1 ij =1 i j S I ↓ ( i j )( n − + k X j =1 ij =1 i j S I ↓ ( i j )( n − = k X j =1 ij =1 i j ( n − m )!( i j − k Y p =1 p = j ( i p )! D m − ,kn + k X j =1 ij =1 i j ( n − m )! k Y p =1 p = j ( i p )! D m − ,k − n − = ( n − m )! k Y j =1 ( i j )! (cid:16) ( k − m ( I )) D m − ,kn − + m ( I ) D m − ,k − n − (cid:17) . (28)We see from Lemma 6.4 that the set of triples ( σ, α, β ) that do not have a common fixed point isin bijective correspondence with the triples in ∪ s,t ∈ [ n ] s = t A [ n ] \{ s,t } (( n − , and ( σ, α, β ) corresponds to (ˆ σ, ˆ α, ˆ β ) where σ = ( s u )( t v )ˆ σ , for some u, v ∈ [ n ] \ { s, t } not necessarily distinct. Of course, theeffect of ( s u ) and ( t v ) is to bring s into the cycle containing u and t into the cycle containing v in ˆ σ . Thus, enumerating the triples ( σ, α, β ) with no common fixed point is equivalent to enumeratingtuples ( s, t, u, v, ˆ σ, ˆ α, ˆ β ) , where s, t ∈ [ n ] are distinct, u, v ∈ [ n ] \ { s, t } are not necessarily distinct,and (ˆ σ, ˆ α, ˆ β ) ∈ A [ n ] \{ s,t } (( n − . Since we are often not concerned with ˆ α and ˆ β (just that they arepart of the relevant triple), we will generally omit writing them. Furthermore, we are trying to determinein this correspondence when σ is I -separated, which will rely on a case analysis on s, t, u, v and on theseparation probabilities of ˆ σ . We discuss this in the three cases given by the size of the intersection of { s, t } with [ m ] .If { s, t } is disjoint from [ m ] , then σ is J -separated if and only if ˆ σ is J -separated in S [ n ] \{ s,t } . Thenumber of quadruples s, t, u, v for which { s, t } are disjoint from [ m ] are ( n − m ) ( n − . Thus, thetotal number of tuples ( s, t, u, v, ˆ σ ) in this case is(29) ( n − m ) ( n − S I ( n − Now suppose that exactly one of s, t ∈ [ m ] , and for now assume it is s (the case for t is the same).Since s ∈ [ m ] , then s ∈ J j for some j . Recall, the effect of ( s u ) on ˆ σ is to bring s into the cycle containing u in ˆ σ . If u / ∈ [ m ] we see that σ is J -separated if ˆ σ is J ( s, u ) -separated, where J ( s, u ) isthe set partition with s replaced by u . The number of such ˆ σ is S I ( n − and the number of quadruples ( s, t, u, v ) when u / ∈ [ m ] is i j ( n − m )( n − m − n − (each factor gives the number of choicesfor s, t, u, v , respectively), giving a total of i j ( n − n − m ) S I ( n − . If u ∈ [ m ] , then u must be inthe same part as s , and σ is J -separated if and only if ˆ σ is J \ { s } -separated. Thus, the number ofchoices of s, t, u, v when u ∈ [ m ] is i j ( n − m )( i j − n − , which gives a total contribution of i j ( i j − n − m )( n − S I ↓ ( i j ) in this case. Summing over all j and multiplying by two to account forthe case t ∈ [ m ] and s / ∈ [ m ] we get(30) n − n − m ) k X j =1 i j S I ( n − + 2( n − m )( n − k X j =1 ( i j ) S I ↓ ( i j )( n − , as the number of tuples ( s, t, u, v, ˆ σ ) where exactly one of s, t ∈ [ m ] .The final case is when both s, t ∈ [ m ] . There are two subcases: u = v and u = v .If u = v , then for σ to be J -separated s, t must be in the same block of J , say J j , as ( s u )( t u ) willbring them into the same cycle of ˆ σ . If u ∈ [ m ] , then u must be in the same block as s and t and σ is J -separated if and only if ˆ σ is J \ { s, t } -separated. There are i j ( i j − i j − choices for s, t and u inthis case. Thus, contributions when u ∈ [ m ] total(31) k X j =1 ( i j ) S I ↓↓ ( i j )( n − , where I ↓↓ ( i j ) is the composition I with i j replaced by i j − . If u / ∈ [ m ] , then σ is J -separated if ˆ σ is J ( s, t ; u ) -separated, where J ( s, t ; u ) is J with s, t replaced by u . Since we have i j ( i j − n − m ) choices for s, t and u , the total contribution u / ∈ [ m ] is(32) k X j =1 ( i j ) ( n − m ) S I ↓ ( i j )( n − . Finally, if u = v , first consider the case where s and t are in the same part of J . Then u and v areeither both in [ m ] , one is in [ m ] and the other is not, or neither is in [ m ] . The contributions in each caseare given by, respectively, a summand of(33) k X j =1 ( i j ) S I ↓↓ ( i j )( n − + 2 k X j =1 ( i j ) ( n − m ) S I ↓ ( i j )( n − + k X j =1 ( i j ) ( n − m ) S I ( n − . If s, t are not in the same part of J , then we have the following cases if σ is J -separated: either u, v ∈ [ m ] , in which case u is in the same block as s and v is in the same block as t in J ; exactly one of u or v is in [ m ] , say u , and then u must be in the same block of J as s ; or, u, v / ∈ [ m ] . The total contributionfrom these cases is(34) k X j =1 k X p =1 p = j ( i j ) ( i p ) S I ↓ ( i j ) ↓ ( i p )( n − +2 k X j =1 X p =1 p = j i j i p ( i j − n − m ) S I ↓ ( i j )( n − + k X j =1 k X p =1 p = j i j i p ( n − m ) S I ( n − . N PRODUCTS OF CYCLES 27
Collecting like terms in (29) - (34), we get(35) ( n − m ) ( n − + k X j =1 n − n − m ) i j + ( i j ) ( n − m ) + i j X p = j i p ( n − m ) S I ( n − + k X j =1 i j ) ( n − m )( n −
2) + ( i j ) ( n − m ) + 2( i j ) ( n − m ) + 2( i j ) X p = j ( i p )( n − m ) S I ↓ ( i j )( n − + k X j =1 (( i j ) + ( i j ) ) S I ↓↓ ( i j )( n − + k X j =1 ( i j ) ( i p ) S I ↓ ( i j ) ↓ ( i p )( n − . The first sum in (35) simplifies as ( n − m ) S I ( n − ( n − + k X j =1 (cid:0) n − i j + ( i j ) + ( i j )( m − i j ) (cid:1) = ( n − m ) S I ( n − ( n − + k X j =1 (2( n − i j + i j m − i j ) = ( n − m ) S I ( n − (cid:0) ( n + m − − m (cid:1) . The second sum above simplifies as ( n − m ) k X j =1 (2( i j ) ( n −
2) + 2( i j ) + 2( i j ) ( m − i j ) + ( i j ) ) S I ↓ ( i j )( n − = ( n − m ) (2 n + 2 m − k X j =1 ( i j ) S I ↓ ( i j )( n − = ( n − m ) (2 n + 2 m − k X j =1 ( i j ) ( n − m − Y x = j ( i x )!( i − D m − ,kn − = ( n − m )! (cid:16)Y ( i x )! (cid:17) (2 n + 2 m − D m − ,kn − k X j =1 ( i j − = ( n − m )! (cid:16)Y ( i x )! (cid:17) (2 n + 2 m − D m − ,kn − ( m − k ) . For the third sum, we are mindful that S I ↓↓ ( i j )( n − is 0 if i j = 1 and I ↓↓ ( i j ) has one less part if i j = 2 .These can both be taken care of with the summation index and by noting that the coefficient is usually 0in such cases, and likewise for the fourth term. Thus, the third and fourth sum in (35) become k X j =1 (cid:0) ( i j − ( i j ) (cid:1) Y x = j ( i x )!( i j − n − m )! D m − ,kn − + k X j =1 X p = j ( i j ) ( i p ) ( n − m )! Y x = p,j ( i x )!( i j − i p − D m − ,kn − = ( n − m )! (cid:16)Y ( i x )! (cid:17) D m − ,kn − X j =1 ij =1 ( i j − + X j =1 ij =1 ( i j − m − i j − k + 1) = ( n − m )! (cid:16)Y ( i x )! (cid:17) D m − ,kn − (( m − k − m − k ) + k − m ( I )) . Combining all the parts we have the following theorem.
Theorem 6.11.
Suppose that n is a positive integer and that I = ( i , . . . , i k ) is a composition of length k of m , with k, m ≤ n . Then the separation probability σ I ( n − , , ( n − , is given by ( n − m )! Q ( i j )! n ( n − (cid:16) ( k − m ( I )) D m − ,kn − + m ( I ) D m − ,k − n − + D m,kn − + (cid:0) ( n + m − − m (cid:1) D m,kn − + ( m − k )(2 n + 2 m − D m − ,kn − + (( m − k − m − k ) + k − m ( I )) D m − ,kn − (cid:17) . In [BDMS14, Equation 1], the authors state that σ Iλ, ( n ) Q ( i j )! depends only on n, λ, m and k . For the specialcase when λ = ( n ) , this is apparent from the explicit form of σ I ( n ) , ( n ) given in (26). In the same spirit,we have the following corollary. Note the unexpected analogy with the case ρ = (1) in Theorem 1.1. Corollary 6.12.
Suppose that n ≥ and I = ( i , . . . , i k ) is a composition of m . Then σ I ( n − , , ( n − , Q ( i j )! depends only on n, m, k and m ( I ) . This result invites us to state the following conjecture. We have computational evidence for thefollowing conjecture for small a and n . Conjecture 6.13.
Let a and n be two non-negative integers with a < n , λ a partition of n and I =( i , . . . , i k ) is a composition of m . Then σ I ( n − a, a ) ,λ Q ( i j )! depends on a, n, λ, m, k, m ( I ) , . . . , m a − ( I ) butnot the higher multiplicities of I . R EFERENCES [BDMS14] O. Bernardi, R. X. Du, A. H. Morales, and R. P. Stanley. Separation probabilities for products of permutations.
Combinatorics, Probability and Computing , pages 201–222, 2014.[Boc80] G. Boccara. Nombres de représentations d’une permutation comme produit de deux cycles de longuers données.
Discrete Math. , 29:105–134, 1980.[CMS12] R. Cori, M. Marcus, and G. Schaeffer. Odd permutations are nicer than even ones.
Eur. J. Comb. , 33(7):1467–1478,2012.[FH59] H. Farahat and G. Higman. The centres of symmetric group rings.
Proc. Roy. Soc. (A) , 250:212–221, 1959.[Fro00] G. Frobenius. Über die Charaktere der symmetrischen Gruppe.
SitzBer. Preuss. Akad. Wiss. Berl. , pages 516–534,1900.[FV12] V. Féray and E. A. Vassilieva. Bijective enumeration of some colored permutations given by the product of twolong cycles.
Discrete Mathematics , 312(2):279 – 292, 2012.[GJ92] I. P. Goulden and D. M. Jackson. The combinatorial relationship between trees, cacti and certain connectioncoefficients for the symmetric group.
European J. Combin. , 13(5):357–365, 1992.[GS98] A. Goupil and G. Schaeffer. Factoring n-cycles and counting maps of given genus.
Eur. J. Comb. , 19(7):819–834,1998.[HSS92] P. Hanlon, J. Stembridge, and R. Stanley. Some combinatorial aspects of the spectra of normally distributed ran-dom matrices.
Comtemporary Mathematics , pages 151–174, 1992.[Jac87] D. M. Jackson. Counting cycles in permutations by group characters, with an application to a topological problem.
Transactions of the American Mathematical Society , 299(2):785–801, 1987.[JV90] D. M. Jackson and T. I. Visentin. A character theoretic approach to embeddings of rooted maps in an orientablesurface of given genus.
Trans. Amer. Math. Soc. , 322(1):343–363, 1990.[Lit50] D. E. Littlewood.
The theory of group characters and matrix representations of groups . American MathematicalSociety, 1950.[LR34] D. E. Littlewood and A. R. Richardson. Group characters and algebra.
Philosophical Trans. Royal Soc. LondonSer. A , 233:99–141, 1934.[Mac95] I. G. Macdonald.
Symmetric functions and Hall polynomials . Oxford Univ P., 2nd edition, 1995.[Sag01] B. Sagan.
The symmetric group : Representations, Combinatorial Algorithms, and Symmetric functions . Springer,New York, second edition, 2001.[Sta81] R. P. Stanley. Factorization of permutations into n -cycles. Discrete Math. , 37:255–262, 1981.[Sta10] R. P. Stanley. Products of cycles. talk given at
Permutation Patterns , slides available at http://math.mit.edu/~rstan/transparencies/cycleprod.pdf , 2010.[Sta11] R. P. Stanley. Two enumerative results on cycles of permutations.
Eur. J. Comb. , 32(6):937–943, 2011.
N PRODUCTS OF CYCLES 29 [Vas12] E. A. Vassilieva. Explicit monomial expansions of the generating series for connection coefficients.
DMTCS proc.of FPSAC , AR:123–134, 2012.[Wal79] D. W. Walkup. How many ways can a permutation be factored into two n-cycles?
Discrete Mathematics , 28(3):315– 319, 1979. A CKNOWLEDGEMENTS
This research project was started during a visit of AR in Bordeaux. This visit was funded via the“invité junior” program of LaBRI. AR would like to thank people in LaBRI for their generous hospitality.V.F. is partially supported by ANR grant PSYCO ANR-11-JS02-001.Both authors would like to thank the anonymous referees for their helpful suggestions. I NSTITUT FÜR M ATHEMATIK , U
NIVERSITÄT Z ÜRICH , Z
ÜRICH , S
WITZERLAND
E-mail address : [email protected] D EPARTMENT OF E CONOMICS , M
ATHEMATICS AND S TATISTICS B IRKBECK C OLLEGE , U
NIVERSITY OF L ONDON M ALET S TREET , L
ONDON
WC1E 7HX, UK
E-mail address ::