Composition of ordinary generating functions
aa r X i v : . [ m a t h . C O ] S e p COMPOSITIONOF ORDINARY GENERATINGFUNCTIONS
Vladimir KruchininTomsk State Universityof Control Systems and RadioelectronicsTomskRussion Federation [email protected]
Abstract
A solution is proposed for the problem of composition of ordinary generating func-tions. A new class of functions that provides a composition of ordinary generatingfunctions is introduced; main theorems are presented; compositae are written for poly-nomials, trigonometric and hyperbolic functions, exponential and log functions. It isshown that the composition holds true for many integer sequences.
Abstract A solution is proposed for the problem of composition of ordinary generatingfunctions. A new class of functions that provides a composition of ordinary generating func-tions is introduced; main theorems are presented; compositae are written for polynomials,trigonometric and hyperbolic functions, exponential and log functions. It is shown that thecomposition holds true for many integer sequences
Generating functions are an efficient tool of solving mathematical problems. Given theordinary generating functions F ( x ) = P n > f ( n ) x n and R ( x ) = P n > r ( n ) x n , the operationof composition of generating functions A ( x ) = R ( F ( x )) is defined correctly. [3, 1, 6, 2].However, coefficients of the composition of generating functions are difficult to find. Stanley[3] came close to the solution of the problem and proposed a formula for the compositionof exponential generating functions based on ordered partitions of a finite set. Let us showthat the basis for the composition of ordinary generating functions is ordered partitions of apositive integer n and put forward basic formulae for the coefficients of the composition ofordinary generating functions. For this purpose, we introduce several definitions.1 efinition 1. An ordinary generating function F ( x ) is a series that belongs to the ring offormal power series in one variable K [[ x ]] : F ( x ) = X n ≥ f ( n ) x n , where f ( n ) : P → K , P is a set of nonnegative numbers; K is a commutative field. Further we consider only ordinary generating functions. The known generating functionsare denoted as F ( x ), R ( x ), G ( x ), and the desired generating function as A ( x ). Definition 2.
An ordered partition (composition) of a positive integer n is an ordered se-quence of positive integers λ i such that k X i =1 λ i = n, where λ i ≥ and k = 1 , n are parts of the ordered partition. C n is a set of all ordered partitions of n . π k ∈ C n is an ordered partition of C n with k parts.The ordered partitions of n have been much studied [4, 5]. Let there be functions f ( n ) and r ( n ) and their generating functions F ( x ) = P n > f ( n ) x n , R ( x ) = P n > r ( n ) x n . Then, calculating the composition of the generating functions A ( x ) = R ( F ( x )) requires [2] [ F ( x )] k = X n ≥ k X λi> λ λ ... + λk = n f ( λ ) f ( λ ) . . . f ( λ k ) x n . (1)Hence it follows that for the function a ( n ) of the composition of generating functions with n >
0, the formula a (0) = r (0) ,a ( n ) = n X k =1 X λi> λ λ ... + λk = n f ( λ ) f ( λ ) . . . f ( λ k ) r ( k ) (2)holds true. Further the composition of generating functions is written implying that a (0) = r (0). Remark 3.
It should be noted that the summation in formulae (1),(2) is over all orderedpartitions of n that have exactly k parts, because { λ + λ + . . . + λ k = n } , λ i > , i = 1 , k (further we use the reduction π k ∈ C n ). 2hus, the ordered partitions of n are the basis for calculation of the composition ofgenerating functions.Let us consider the following example. Assume that f (0) = 0, f ( n ) = 1 for all n > F ( x ) = x − x . Then, the expression X π k ∈ C n f ( λ ) f ( λ ) . . . f ( λ k )gives the number of ordered partitions of n with exactly k parts; this number is equal to (cid:0) n − k − (cid:1) [4]. Thus, X π k ∈ C n f ( λ ) f ( λ ) . . . f ( λ k ) = (cid:18) n − k − (cid:19) . Hence it follows that the formula valid for any generating function R ( x ) = P n > r ( n ) x n and A ( x ) = R (cid:0) x − x (cid:1) is a ( n ) = n X k =1 (cid:18) n − k − (cid:19) r ( k ) . Example 4.
For R ( x ) = x − x , we have the composition A ( x ) = x − x and a ( n ) = n X k =1 (cid:18) n − k − (cid:19) = 2 n − . Thus, we calculate the total number of ordered partitions of n . Example 5.
We have R ( x ) = e x , then for the composition A ( x ) = e x − x we can write a ( n ) = n X k =1 (cid:18) n − k − (cid:19) k ! (see A000262 formula Herbert S. Wilf ). Example 6.
We have R ( x ) = x − x − x , then for the composition A ( x ) = R ( x − x ) we can write a ( n ) = n X k =1 (cid:18) n − k − (cid:19) F ( k ) , where F ( k ) is the Fibonacci numbers (see A001519, formula Benoit Cloitre). Definition 7.
A composita of the generating function F ( x ) = P n> f ( n ) x n is the function F ∆ ( n, k ) = X π k ∈ C n f ( λ ) f ( λ ) . . . f ( λ k ) . (3)3alculation of F ∆ ( n, k ) is of prime importance to obtain a composition of generatingfunctions, because from formula (2) it follows that the formula valid for the composition A ( x ) = R ( F ( x )) is a ( n ) = n X k =1 F ∆ ( n, k ) r ( k ) . (4)The basis for the derivation of a composita is calculation of the ordered partition π k of C n .From formula (1) it follows that the generating function of the composita is equal to[ F ( x )] k = X n ≥ k F ∆ ( n, k ) x n . For F ( x ), the condition f (0) = 0 holds true, and hence numbering for the composita beginswith k = 1, n = 1. For k = 1, F ∆ ( n, k ) = f ( n ). For k > n , F ∆ ( n, k ) is equal to zero. Thisstatement stems from the fact that there is no ordered partition of n in which the numberof parts is larger than n .The above example demonstrates that the Pascal triangle is a composita for the gener-ating function x − x and deriving the composition A ( x ) = R (cid:0) x − x (cid:1) requires the use of F ∆ ( n, k ) = (cid:18) n − k − (cid:19) . Let us derive a recurrence formula for the composita of a generating function.
Theorem 8.
For the composita F ∆ ( n, k ) of the generating function F ( x ) = P n> f ( n ) x n ,the following relation holds true: F ∆ ( n, k ) = f ( n ) , k = 1 , [ f (1)] n , k = n, P n − ki =0 f ( i + 1) F ∆ ( n − i − , k − k < n. (5) Proof.
Let us derive a recurrence formula for the c n,k number of ordered partitions of n thathave exactly k parts. Let us introduce the operation pos [ λ ∗ , π k ] of adjunction of the new part λ ∗ on the left to a certain ordered partition π k ∈ C n providing that λ ∗ >
0. From the orderedpartition π k ∈ C n this operation obtains an ordered partition π k +1 ∈ C λ ∗ + n . Let us extendthis operation to sets. Assume that C n,k = { π k | π k ∈ C n } , then the set ˆ C = pos [ λ ∗ , C n,k ] is asubset C λ ∗ + n,k +1 . Thus, we can write C n,k = pos [1 , C n − ,k − ] ∪ pos [2 , C n − ,k − ] ∪ . . . ∪ pos [ n − k − , C k − ,k − ] . In this case, the condition pos [ i, C n − i,k − ] ∩ pos [ j, C n − j,k − ] = ⊘ is fulfilled for all i = j , because the first parts of the ordered partitions π k do not coincide.Hence, c n,k = n − k X i =0 c n − i − ,k − , (6)4nd c k,k = 1 because we have the only ordered partition π k = { . . . + 1 = n } , and c n, = 1 because π = { n = n } .Let us now consider expression (3). Using expression (6), we can write F ∆ ( n, k ) == f (1) F ∆ ( n − , k −
1) + f (2) F ∆ ( n − , k −
1) + . . . ++ f ( n − k + 1) F ∆ ( k − , k − . The set C n,n consists of the only ordered partition { . . . + 1 } , and then F ∆ n,n = [ f (1)] n ;the set C n, consists of { n } , and then F ∆ n, = f ( n ). Thus, the theorem is proved.Consideration of formula (4) allows the conclusion that the composita does not dependon R ( x ) and characterizes the generating function F ( x ). In tabular form, the composita isrepresented as F ∆1 , F ∆2 , F ∆2 , F ∆3 , F ∆3 , F ∆3 , F ∆4 , F ∆4 , F ∆4 , F ∆4 , . . . ... ... ... . . . F ∆ n, F ∆ n, . . . . . . F ∆ n,n − F ∆ n,n or, knowing that F ∆1 ,n = f ( n ), F ∆ n,n = [ f (1)] n , as f (1) f (2) f (1) f (3) F ∆3 , f (1) f (4) F ∆4 , F ∆4 , f (1). . . ... ... ... . . . f ( n ) F ∆ n, . . . . . . F ∆ n,n − f n (1)Below are the terms of the composite of the generating function F ( x ) = x − x :11 11 2 11 3 3 11 4 6 4 11 5 10 10 5 1 Theorem 9.
For a given ordinary generating function F ( x ) = P n ≥ f ( n ) x n , the composita F ∆ ( n, k ) always exists and is unique.Proof. Without proof. 5
Calculation of compositae
Calculation of compositae is based on derivation of the generating function of a composita[ A ( x )] k = X n > k A ∆ ( n, k ) x n and operation on them. Theorem 10.
Let there be a generating function F ( x ) = P n> f ( n ) x n , its composita F ∆ ( n, k ) , and a constant α . Then, the generating function A ( x ) = αF ( x ) has the com-posita A ∆ ( n, k ) = α k F ∆ ( n, k ) . Proof. [ A ( x )] k = [ αF ( x )] k = α k [ F ( x )] k . Theorem 11.
Let there be a generating function F ( x ) = P n> f ( n ) x n , its composita F ∆ ( n, k ) , and a constant α . Then, the generating function A ( x ) = F ( αx ) has the com-posita A ∆ ( n, k ) = α n F ∆ ( n, k ) . Proof.
By definition, we have A ∆ ( n, k ) = X π k ∈ C n α λ f ( λ ) α λ f ( λ ) . . . α λ k f ( λ k ) == α n X π k ∈ C n f ( λ ) f ( λ ) . . . f ( λ k ) = α n F ∆ ( n, k ) . Theorem 12.
Let there be a generating function F ( x ) = P n> f ( n ) x n , its composita F ∆ ( n, k ) , a generating function B ( x ) = P n > b ( n ) x n and [ B ( x ) k ] = P n > B ( n, k ) x n . Then,the generating function A ( x ) = F ( x ) B ( x ) has the composita A ∆ ( n, k ) = n X i = k F ∆ ( i, k ) B ( n − i, k ) . Proof.
Because a (0) = f (0) b (0) = 0, A ( x ) has the composita A ∆ ( n, k ). On the other hand,[ A ( x )] k = [ F ( x )] k [ B ( x )] k . This, reasoning from the rule of product of generating functions, gives A ∆ ( n, k ) = n X i = k F ∆ ( i, k ) B ( n − i, k ) . B ( x ) b (0) = 0, the formula has the form: A ∆ ( n, k ) = n − k X i = k F ∆ ( i, k ) B ∆ ( n − i, k ) . Theorem 13.
Let there be generating functions F ( x ) = P n> f ( n ) x n , G ( x ) = P n> g ( n ) x n and their compositae F ∆ ( n, k ) , G ∆ ( n, k ) . Then, the generating function A ( x ) = F ( x )+ G ( x ) has the composita A ∆ ( n, k ) = F ∆ ( n, k ) + k − X j =1 (cid:18) kj (cid:19) n − k + j X i = j F ∆ ( i, j ) G ∆ ( n − i, k − j ) + G ∆ ( n, k ) . Proof.
According to the binomial theorem, we have[ A ( x )] k = k X j =0 (cid:18) kj (cid:19) [ F ( x )] j [ G ( x )] k − j , [ F ( x )] j = X n > j F ∆ ( n, j ) , and [ G ( x )] k − j = X n > k − j G ∆ ( n, k − j ) . According to the rule of multiplication of series, we obtain A ∆ ( n, k ) = F ∆ ( n, k ) + k − X j =1 (cid:18) kj (cid:19) n − k + j X i = j F ∆ ( i, j ) G ∆ ( n − i, k − j ) + G ∆ ( n, k ) . Definition 14.
Let there be a composition of generating functions A ( x ) = R ( F ( x )) . Then,the product of two compositae will be termed a composite of the composition A ( x ) and denotedas A ∆ ( n, k ) = F ∆ ( n, k ) ◦ R ∆ ( n, k ) . Theorem 15.
Let there be two generating functions F ( x ) = P n> f ( n ) x n and R ( x ) = P n> r ( n ) x n , and their compositae F ∆ ( n, k ) and R ∆ ( n, k ) . Then, the expression valid forthe product of the compositae A ∆ = F ∆ ◦ R ∆ is A ∆ ( n, m ) = n X k = m F ∆ ( n, k ) R ∆ ( k, m ) . (7) Proof. [ A ( x )] m = [ G ( F ( x )] m = G m ( F ( x )) . Hence, according to the composition rule and taking into account that the nonzero terms G ∆ ( n, m ) begin with n > m , we have A ∆ ( n, m ) = n X k = m F ∆ ( n, k ) G ∆ ( k, m ) . orollary . Because the composition of generating functions is an associative operationand F ( x ) ◦ ( R ( x ) ◦ G ( x )) = ( F ( x ) ◦ R ( x )) ◦ G ( x ) , the product of compositae is also an associative operation and n X k = m n X i = k F ∆ ( n, i ) R ∆ ( i, k ) G ∆ ( k, m ) = n X k = m n X i = k R ∆ ( n, i ) G ∆ ( i, k ) F ∆ ( k, m ) . Definition 16.
An identical composita Id ∆ ( n, k ) is a composita of the generating function F ( x ) = x . By definition, [ F ( x )] k = x k . Then F ∆ ( n, k ) = (cid:26) , n = k, , n = k. (8)Thus, F ∆ ( n, k ) = δ n,k , where δ n,k is the Kronecker delta. It is easily seen that for anygenerating function A ( x ), we have the identity a ( n ) = n X k =1 Id ∆ ( n, k ) a ( k ) . The composita of the function F ( x ) = x m is F ∆ ( n, k, m ) = δ nm ,k , mod ( n, m ) = 0 or n = km. (9) P ( x ) = ( ax + bx )Let us consider P ( x ) = ( ax + bx ). Then, p (0) = 0, p (1) = a and p (2) = b , and the restare p ( n ) = 0 , n >
2. The composita of the function F ( x ) = ax is equal to a k δ n,k , and thecomposita of the function G ( x ) = bx is equal to b k δ n ,k . Using sum theorem (13), we obtain P ∆2 ( n, k ) = k X j =0 (cid:18) kj (cid:19) n − k + j X i = j a j δ i,j b k − j δ n − i ,k − j ,δ n − i ,k − j = 1 for n − i = k − j , whence i = n − k + 2 j . So we have P ∆2 ( n, k ) = k X j =0 (cid:18) kj (cid:19) a j δ n − k +2 j,j b k − j . δ n − k +2 j,j = 1 for n − k + 2 j = j , whence j = 2 k − n . So we obtain P ∆2 ( n, k, a, b ) = (cid:18) kn − k (cid:19) a k − n b n − k (10)for ⌈ n ⌉ ≤ k ≤ n .Thus, the composition A ( x ) = R ( ax + bx ) can be found using the expression: a ( n ) = n X k = ⌈ n ⌉ (cid:18) kn − k (cid:19) a k − n b n − k r ( k ) . For example, let us derive an expression for the coefficients of the generating function A ( x ) = e x + x (see A000085 ). Taking into account that this function is an exponential generatingfunction, we obtain a ( n ) = n ! n X k = ⌈ n ⌉ (cid:18) kn − k (cid:19) n − k k ! . Another example is A ( x ) = R ( F ( x )), where R ( x ) = x − x and F ( x ) = x + x , A ( x ) = x + x − x − x .Hence a ( n ) = n X k = ⌈ n ⌉ (cid:18) kn − k (cid:19) (see A000045). P ( x ) = ax + bx + cx The polynomial P ( x ) = ax + bx + cx can be expressed as P ( x ) = ax + xP ( x, b, c ) . The composita ax is equal to δ ( n, k ) a k , and the composita xP ( x ) to A ∆( n − k, k ). Then,on the strength of the theorem on the composita of the sum of generating functions, we have A ∆3 ( n, k, a, b, c ) = k X j =0 (cid:18) kj (cid:19) n − k + j X i = j A ∆( i − j, j, b, c ) δ ( n − i, k − j ) a k − j . Simplification gives δ ( n − i, k − j ) = 1 for n − i = k − j , whence we have i = n − k + j and A ∆3 ( n, k, a, b, c ) = k X j =0 (cid:18) kj (cid:19) A ( n − k, j, b, c ) a k − j , where A ∆2 ( n − k, j, b, c ) = (cid:0) jn − k − j (cid:1) b j + k − n b n − k − j . Hence, A ∆3 ( n, k, a, b, c ) = k X j =0 (cid:18) kj (cid:19)(cid:18) jn − k − j (cid:19) a k − j b j + k − n b n − k − j . A ( x ) = − ax − bx − cx , the following expression holds true: a ( n ) = n X k =1 k X j =0 (cid:18) kj (cid:19)(cid:18) jn − k − j (cid:19) a k − j b j + k − n b n − k − j . P ( x ) = ax + cx An important condition in the foregoing examples is that a, b, c = 0. Therefore, if b = 0 theformula for the composita should be sought for over again. For example, P ( x ) = ax + cx . In this case, the expression for the composita is P ∆ ( n, k ) = (cid:18) k k − n (cid:19) a k − n c n − k , where ( n − k ) is exactly divisible by 2. For example, for the generating function A ( x ) = − x − x the following expression holds true: a ( n ) = n X k =1 (cid:18) k k − n (cid:19) (see A000930). P ( x ) = ax + bx + cx + dx At n = 4, the polynomial P ( x ) = ax + bx + cx + dx can be expressed as P ( x ) = P ( x ) + x P ( x ) . The generating function of the composita for x P ( x ) is equal to x k (cid:18) kn − k (cid:19) c k − n b n − k x n = (cid:18) kn − k (cid:19) c k − n b n − k x n +2 k , and hence the expression for the composita is (cid:18) kn − k (cid:19) c k − n b n − k . Then the composita P ( x ) has the following expression: k X j =0 (cid:18) kj (cid:19) n − k + j X i = j (cid:18) ji − j (cid:19) a j − i b i − j (cid:18) k − jn − i − k − j ) (cid:19) c k − j ) − ( n − i ) d n − i − k + j . A ( x ) = − ax − bx − cx − dx the following expressionholds true: a ( n ) = n X k =1 k X j =0 (cid:18) kj (cid:19) n − k + j X i = j (cid:18) ji − j (cid:19) a j − i b i − j (cid:18) k − jn − i − k − j ) (cid:19) c k − j ) − ( n − i ) d n − i − k + j . At a = b = c = d = 1, we obtain the generating function A ( x ) = − x − x − x − x . Hence a ( n ) = n X k =1 k X j =0 (cid:18) kj (cid:19) n − k + j X i = j (cid:18) ji − j (cid:19)(cid:18) k − jn − i − k − j ) (cid:19) . P ( x ) = ax + bx + cx + dx + ex For finding the composita of the m th power polynomial, we can propose the recurrent algo-rithm A ∆ m ( n, k ) = k X j =0 A ∆ m − ( n − k, j ) a k − j , providing that A m − ( n,
0) = 1. Using this recurrent algorithm, we obtain the composita ofthe 5th power polynomial: k X r =0 a k − r (cid:18) kr (cid:19) r X m =0 b r − m m X j =0 c m − j d r − n + m + k +2 j e v (cid:18) jv (cid:19) (cid:18) mj (cid:19)! (cid:18) rm (cid:19) , where v = − r + n − m − k − j . A ( x ) = ( ax − bx ) For the generating function F ( x ) = x (1 − x ) , F ∆ ( n, k ) = (cid:0) n − k − (cid:1) , and A ( x ) = ( ab − bx − bx ) . Using theorems (10,11), we obtain A ∆ ( n, k ) = (cid:18) n − k − (cid:19) a k b n − k . Let us find the expression for the coefficients of the generating function [ B ( x )] k = e kx : B ( x ) k = e xk = X n > k n n ! , B ( n, k ) = k n n ! . Now, for A ( x ) = xe x the composita is equal to A ∆ ( n, k ) = B ( n − k, k ) = k n − k ( n − k )! . (11)Let us write the composita for the generating function A ( x ) = e x − A ( x ) k = k X m =0 (cid:18) km (cid:19) e mx ( − k − m , whence it follows that the composita is A ∆ ( n, k ) = k X m =0 (cid:18) km (cid:19) m n n ! ( − k − m = k ! n ! S ( n, k ) , (12)where S ( n, k ) is the Stirling numbers of the second kind. For the generating functions ofthe Bell numbers A ( x ) = e e x − , we have a ( n ) = n ! n X k =1 S ( n, k ) k ! n ! 1 k ! = n X k =1 S ( n, k )(see A000110). ln(1 + x ) Let F ( x ) = ln( x + 1). Then, knowing the relation [6] ∞ X n = k S ( n, k ) x n n ! = [ln(1 + x )] k k ! , where S ( n, k ) is the Stirling numbers of the first kind, and using formula (1), we obtain theexpression for the composita of the generating function ln(1 + x ): F ∆ ( n, k ) = k ! n ! S ( n, k ) . (13) The generating function of the Bernoulli numbers is A ( x ) = xe x − . B ( F ( x )), where B ( x ) = ln xx , F ( x ) = e x −
1. Let us find the expression for the coefficients of the generating function [ B ( x )] k :[ B ( x )] k = X n > S ( n, k ) k ! n ! x n − k , whence B ( n, k ) = S ( n + k, k ) k !( n + k )! . Knowing the composita of the function F ( x ) (see 12), F ∆ ( n, k ) = k ! n ! S ( n, k ) . Let us write the composition of the generating functions A ( x ) k = [ B ( e x − k : A ( n, m ) = (cid:26) , n = 0 , P nk =1 S ( n, k ) k ! n ! S ( k + m, m ) m !( k + m )! , n > . Then the composita of xA ( x ) is A ∆ ( n, m ) = (cid:26) , n = m, m !( n − m )! P n − mk =1 k !( k + m )! S ( k + m, m ) S ( n − m, k ) , n > m. Let us find the composita for the generating function of the Fibonacci numbers: A ( x ) = x − x − x . The function can be represented as the composition of the generating functions A ( x ) = R ( F ( x )), where R ( x ) = x − x , F ( x ) = x − x . Let us find the composita for F ( x ): F ∆ ( n, k ) = ( (cid:0) n + k − k − (cid:1) , at n + k – even , , at n + k – odd . Now, using the operation of product of compositae, we find the composita of the generatingfunction A ( x ): A ∆ ( n, m ) = n P k = m (cid:0) n + k − k − (cid:1)(cid:0) k − m − (cid:1) , at n + k – even . Below are the first terms of the composita for the generating function of the Fibonaccinumbers: 11 12 2 13 5 3 15 10 9 4 18 20 22 14 5 113 38 51 40 20 6 113 .8 Composita for the generalized Fibonacci numbers
Let us find the composita of the generating function: F ( x ) = x + x + . . . + x m = x − x m +1 − x . Let us write F ( x ) as the product of the functions G ( x ) = x − x m +1 and R ( x ) = − x .Let us find the composita for G ( x ). For this purpose, we consider the compositae of thefunctions y ( x ) = x and z ( x ) = − x m . For y ( x ), the composita is equal to Id ( n, k ) = δ n,k .For z ( x ) = − x m , the composita is Z ∆ ( n, k ) = ( − k δ nm ,k . Then, on the strength of the theorem on the composite of sum of generating functions y ( x ) + z ( x ), we have G ∆ ( n, k ) = X j =0 (cid:18) kj (cid:19) n − k + j X i = j Id ( i, j ) Z ∆ ( n − i, k − j ) == X j =0 (cid:18) kj (cid:19) n − k + j X i = j δ i,j δ n − im ,k − j ( − k − j . The function δ i,j = 1 is only for i = j , and hence G ∆ ( n, k ) = X j =0 (cid:18) kj (cid:19) δ n − jm ,k − j ( − k − j . The function δ n − jm ,k − j = 1 is only for n − jm = k − j , and hence G ∆ ( n, k ) = (cid:18) k ( m +1) k − nm (cid:19) ( − n − km . It is known that R ( n, k ) = (cid:0) n + k − k − (cid:1) . Then, with regard to the rule of finding the compositaof the product of generating functions (case 2), we obtain F ∆ ( n, k ) = n X i = k (cid:18) k ( m +1) k − im (cid:19) ( − i − km (cid:18) n − i + k − k − (cid:19) . Let us consider the composita of the generating functions: A ( x ) = F ( x )1 − F ( x ) = x − x m +1 − x − x m +1 . A ( x ): A ∆ ( n, l ) = n X k = l F ∆ ( n, k ) (cid:18) k − m − (cid:19) == n X k = m n X i = k (cid:18) k ( m +1) k − im (cid:19) ( − i − km (cid:18) n − i + k − k − (cid:19)(cid:18) k − l − (cid:19) . For l = 1, we derive the formula for the generalized Fibonacci numbers: F ( m ) n = n X k =1 n X i = k (cid:18) k ( m +1) k − im (cid:19) ( − i − km (cid:18) n − i + k − k − (cid:19) . (14) Let F ( x ) = x −√ − x x , then the composita has the form F ∆ ( n, k ) = n − k X i =0 C ( i ) F ∆ n − i − ,k − , where C ( i ) is the Catalan numbers. The composita F ∆ ( n, k ) has the following triangularform: 11 12 1 15 5 3 114 14 9 4 1Let us consider the sequence A009766 called the Catalan triangle. This triangle is given bythe formula a ( n, m ) = (cid:18) n + mn (cid:19) n − k + 1 n + 1 . Below are the initial values of the triangle, and n and m begin with zero.11 12 1 21 3 5 51 4 9 14 14Comparison of two triangles suggests that a ( n, k ) = F ∆ ( n + 1 , n − k + 1). Hence, thecomposita for the Catalan generating function is equal to F ∆ ( n, k ) = (cid:18) n − k − n − (cid:19) kn . A ( x ) = R ( −√ − x ) is a ( n ) = (cid:18) n − k − n − (cid:19) kn · r ( k ) . x √ − x This generating function can be represented as the composition of the functions: x √ − x = x − (cid:18) √ − x − (cid:19) = x − C ( x ) , where C ( x ) = −√ − x .Using the formula of composition, we finally obtain A ∆ ( n, m ) = (cid:26) , n = m, P n − mk =1 (cid:0) n − m − k − n − m − (cid:1) kn − m k − n +2 m (cid:0) k + m − m − (cid:1) , n > m. Using the expression sin( x ) = e ix − e − ix i , we obtain sin( x ) k :sin( x ) k = 12 k i k k X m =0 (cid:18) km (cid:19) e imx e − i ( k − m ) x ( − k − m = 12 k i k k X m =0 (cid:18) km (cid:19) e i (2 m − k ) x ( − k − m . Hence the composita is 12 k i n − k k X m =0 (cid:18) km (cid:19) (2 m − k ) n n ! ( − k − m . Taking into account that n − k is an even number and the function is symmetric about k , we obtain the composita of the generating function sin( x ): A ∆ ( n, k ) = ( k − n ! P ⌊ k ⌋ m =0 (cid:0) km (cid:1) (2 m − k ) n ( − n − k − m , ( n − k ) − even0 , ( n − k ) − odd Example 17.
For the Euler numbers we know the exponential generating function − sin ( x ) .Hence, E n +1 = n X k =1 n + k even k − ⌊ k ⌋ X m =0 (cid:18) km (cid:19) (2 m − k ) n ( − n + k − m (see A000111). xample 18. For the generating function A ( x ) = e sin( x ) , the valid expression is a n = n X k =1 n + k even k − k ! ⌊ k ⌋ X m =0 (cid:18) km (cid:19) (2 m − k ) n ( − n + k − m (see A002017). Knowing that cos( x ) = e ix + e − ix , We have [cos x ] k = 12 k k X j =0 (cid:18) kj (cid:19) e (2 j − k ) ix == 12 k X n > k X j =0 (cid:18) kj (cid:19) (2 j − k ) n i n x n n ! . Hence B ( n, k ) = 12 k n ! ( − n k X j =0 (cid:18) kj (cid:19) (2 j − k ) n . Then, the composita of the generating function xcos ( x ) is A ∆ ( n, k ) = ( k − ( n − k )! ( − n − k P kj =0 (cid:0) kj (cid:1) (2 j − k ) n − k , n − k − even0 , n − k − odd . The composita of the function cos ( x ) − A ∆ ( n, k ) = k X i =0 B ( n, i )( − k − i . Let us consider the following example. Let there be a generating function A ( x ) = sec ( x ) = cos ( x ) = x ) − . Hence, on the strength of the formula of composition andcomposita (cos( x ) − a ( n ) = n X k =1 k X m =0 (cid:18) km (cid:19) − m m X j =0 (2 j − m ) n (cid:18) mj (cid:19) ( − n + m (see A000364). 17 .11.3 Composita for tan( x )For the tangent, we know the identitytan( x ) = e ix − e − ix i ( e ix − e − ix ) . Division of the numerator and denominator by e ix givestan( x ) = 1 − e − ix i (1 − e − ix ) . Multiplication of the numerator and denominator by i, and addition and then subtractionof unity gives tan( x ) = i e − ix − − ( e − ix − . Whence it follows that tan( x ) = i e − ix − − ( e − ix − . Thus, the function tan( x ) is expressed as the composition of the functions F ( x ) = i x x and functions R ( x ) = e − ix −
1. The composita for F ( x ) is equal to F ∆ ( n, k ) = 12 n ( − n − k (cid:18) n − k − (cid:19) i k . The composita for R ( x ) is equal to R ∆ ( n, k ) = ( − i ) n k ! n ! S ( n, k ) , where S ( n, k ) is the Stirling numbers of the second kind. Then, on the strength of thetheorem on the product of compositae, we obtain the composita of the function tan( x ): A ∆ ( n, m ) = n X k = m ( − i ) n S ( n, k ) k ! n ! 12 k ( − k − m (cid:18) k − m − (cid:19) i m . After transformation, we obtain A ∆ ( n, m ) = ( − n + m n X k = m (2) n − k S ( n, k ) k ! n ! ( − n + k − m (cid:18) k − m − (cid:19) . Then at k = 1, the expression for the tangential numbers is a ( n ) = ( − n +1 2 n +1 X j =1 ( − j j ! 2 n − j +1 S (2 n + 1 , j )18see A000182)Let us consider the example A ( x ) = e tan( x ) : a ( n ) = n X k =1 ( − n + k P nj = k (cid:0) j − k − (cid:1) j ! 2 n − j ( − n − k + j S ( n, j ) k !(see A006229). For more examples, see A000828,A000831,A003707 x cot( x )It is known that x cot( x ) = ix e ix + e − ix e ix − e − ix = ix + 2 ix e ix − . The composita ix is equal to δ ( n , k ) i k , and the composita for ix e ix − is equal to(2 i ) n − k B ∆ ( n, k ) , where B ∆ ( n, k ) is the composita for the generating function of the Bernoulli numbers. Usingthe theorem on the composita of the sum of generating functions, we obtain the compositaof the function x cot( x ): A ∆ ( n, k ) = δ ( n , k ) i k + k X j =1 B ∆ ( n − k + 2 j, j )(2 i ) n − k + j i k − j == δ ( n , k ) i k + i n − k k X j =1 B ∆ ( n − k + 2 j, j )2 n − k + j F ( x ) = arctan( x )Let us consider the generating function of the arc tangent:arctan( x ) = X n ≥ ( − n (2 n + 1) x n +1 . Let us find an expression for the composita of the arc tangent from the operation of productof compositae. For this purpose, the expressionarctan( x ) = i − ix )) − ln(1 + ix ))is written as follows: arctan( x ) = i − ix ix ) . The composita of the function f ( x ) = ix ix is equal to F ∆ ( n, k ) = 2 k (cid:18) n − k − (cid:19) i n , A ∆ z ( n, m ) = i m m n X k = m k (cid:18) n − k − (cid:19) i n m ! k ! S ( k, m ) . (15) A ∆ z ( n, m ) = ( − m + n m n X k = m k (cid:18) n − k − (cid:19) m ! k ! S ( k, m ) . (16)Below are the first terms of the composita of the arc tangent A ∆ ( n, k ) in the triangularform: 10 1 − − − − − − Example 19.
Let there be R ( x ) = − x , then the coefficients of the generating function A ( x ) = 11 − arctan( x ) are expressed by the formula: a ( n ) = n X m =1 ( − m + n m n X k = m k (cid:18) n − k − (cid:19) m ! k ! S ( k, m ) . Hence, summation of rows of the composita of the arc tangent gives the following series: A ( x ) = 1 + x + x + 23 x + 13 x + 15 x + 845 x + . . . . Example 20.
Let A ( x ) = e arctan( x ) , then the valid expression is a ( n ) = n ! n X m =1 ( − n + m P ni = m i S ( i,m ) ( n − i − ) i ! m (see A002019). For the hyperbolic sine, we have the known expression:sinh( x ) = e x − e − x . (cid:18) e x − e − x (cid:19) k = 12 k (cid:0) e x + e − x (cid:1) k = 12 k k X i =0 (cid:18) ki (cid:19) e ( k − i ) x ( − i e − ix == 12 k k X i =0 ( − i (cid:18) ki (cid:19) e ( k − i ) x . Let us write e x as a series, then we obtain12 k k X i =0 ( − i (cid:18) ki (cid:19) X n > ( k − i ) n n ! x n . Hence, the composita is F ∆ ( n, k ) = 12 k k X i =0 ( − i (cid:18) ki (cid:19) ( k − i ) n n ! . For example, for A ( x ) = e sinh x the valid expression is a ( n ) = n X k =1 P ki =0 ( − i ( k − i ) n (cid:0) ki (cid:1) k k !(see A002724).For the hyperbolic cosine, we havecosh( x ) = e x + e − x . Then, cosh k ( x ) = (cid:18) e x + e − x (cid:19) k = 12 k k X i =0 (cid:18) ki (cid:19) e ( k − i ) x == 12 k k X i =0 (cid:18) ki (cid:19) X n > ( k − i ) n n ! x n , and hence the composita of the generating function x cosh( x ) is F ∆ ( n, k ) = 12 k k X i =0 (cid:18) ki (cid:19) ( k − i ) n − k ( n − k )! . For example, for A ( x ) = e cosh x the valid expression is n X k =1 (cid:16)P ki =0 ( k − i ) n − k (cid:0) ki (cid:1)(cid:17) (cid:0) nk (cid:1) k see A003727). 21 Conclusion
The operation of the composition A ( x ) = R ( F ( x )) of ordinary generating functions requires:1. Finding the composita F ∆ ( n, k ) of the generating function F ( x ) with the use oftheorems (10,11,12,13,15)2. Writing the composition in the form a ( n ) = n X k =1 F ∆ ( n, k ) r ( n ) . References [1] A. I. Markushevich, ”Theory of functions of a complex variable” , 1 , Chelsea (1977)(Translated from Russian)[2] G. P. Egorichev, ”Integral representation and the computation of combinatorial sums”, Amer. Math. Soc. (1984) (Translated from Russian)[3] R. P. Stanley.
Enumerative combinatorics
2, volume 62 of
Cambridge Studies in Ad-vanced Mathematics . Cambridge University Press, Cambridge, 1999.[4] G. E. Andrews.
Number Theory , W. B. Sounders Company, 1981.[5] V. V. Kruchinin.
Combinatorics of Compositions and Aplications , V-Spektr, Tomsk,2010. (in rus)[6] R. L. Graham, D. E. Knuth, and O. Patashnik,
Concrete mathematics , Addison-Wesley,Reading, MA, 1989.2000
Mathematics Subject Classification : Primary 05A15; Secondary 30B10.