Solutions for the problems of problem set 3 1. 4 1 4 8 1 2 Partial Young-tableauxs (P1 , Q1 ) and (P2 , Q2 ). 3 8 4 1 8 3 4 1 2 3 1 2 3 4 Partial Young-tableauxs (P3 , Q3 ) and (P4 , Q4 ). 1 7 3 8 4 1 2 3 5 4 1 7 11 3 8 4 1 2 6 3 5 4 Partial Young-tableauxs (P5 , Q5 ) and (P6 , Q6 ). 1 5 11 3 7 4 8 1 2 6 3 5 4 7 1 5 6 3 7 11 4 8 1 2 6 3 5 8 4 7 Partial Young-tableauxs (P7 , Q7 ) and (P8 , Q8 ). 1 5 6 9 3 7 11 4 8 1 2 6 9 3 5 11 4 7 8 1 2 6 9 3 5 8 4 7 1 2 6 9 3 5 8 4 7 10 Partial Young-tableauxs (P9 , Q9 ) and (P10 , Q10 ). 1 5 6 9 12 3 5 11 4 7 8 1 2 6 9 11 3 5 8 4 7 10 1 2 6 9 10 1 2 6 9 11 3 5 11 12 3 5 8 12 4 7 4 7 8 10 Partial Young-tableauxs (P11 , Q11 ) and (P12 , Q12 ). 1 2 2. We prove the statement by induction on n. For n = 1 the statement is trivial. Given a permutation π ∈ Sn+1 , then deleting n + 1 from it we get a permutation π 0 ∈ Sn . Then I(π) = I(π 0 ) + (n + 1 − π(n + 1)). Clearly, if we have a π 0 ∈ Sn we can insert n + 1 into n + 1 places thus obtaining a permutation π with I(π) = I(π 0 ) + n + 1 − k where k = π(n + 1) runs between 0 and n. Then n+1 X X X X 0 I(π) I(π 0 ) q = q q n+1−k = (1 + q + · · · + q n ) q I(π ) π 0 ∈Sn π∈Sn+1 k=1 π 0 ∈Sn By induction this is equal to (1 + q)(1 + q + q 2 ) . . . (1 + q + q 2 + · · · + q n−1 )(1 + q + · · · + q n ) = (n + 1)q !. 3. We prove the statement by induction on n. The claim is trivial for n = 1. We will use that n+1 n n n+1−k = +q . k k q k−1 q q Hence ! n+1 X k n + 1 n n q( ) q (2) tk = + q n+1−k tk = k k k − 1 q q q k=0 k=0 n+1 X k 2 n+1 X k−1 k n n k n ( ) ( ) 2 2 tk−1 = t +q t q = q k − 1 k q q k=1 k=0 n n X k n X k n tk = tk + q n t q (2) = q (2) k q k q k=0 k=0 n X = (1 + t)(1 + qt) . . . (1 + q n−1 t) + q n t(1 + t)(1 + qt) . . . (1 + q n−1 t) = = (1 + t)(1 + qt) . . . (1 + q n−1 t)(1 + q n t). n n k k−1 We used that n+1 = = 0 and + n + 1 − k = + n. −1 q 2 2 q 4. (a) Assume that given a partition of n with exactly t pieces of 1’s, then after erasing the 1’s one by one, we get a partition of n − 1, n − 2, . . . , n − t. And of course, given a partition of some m we can add n − m 1’s to get a partition of n. Hence sn (1) = p(n − 1) + p(n − 2) + · · · + p(0). (b) The analogous argument shows that sn (2) = p(n − 2) + p(n − 4) + · · · + p(n−2bn/2c). And in general, sn (t) = p(n−t)+p(n−2t)+· · ·+p(n−tbn/tc). 3 (c) Note that np(n) = n X isn (i), i=1 since both sides counts the sum of numbers in all partitions of n. Then np(n) = n X i (p(n − i) + p(n − 2i) + · · · + p(n − ibn/ic)) . i=1 Now n X i (p(n − i) + p(n − 2i) + · · · + p(n − ibn/ic)) = i=1 n X σ(k)p(n − k) k=1 since a term p(n − k) appears at every divisor of k. Hence np(n) = n X σ(k)p(n − k). k=1 5. Let λ = (λ1 , . . . , λm ) be a partition of n, so λ1 ≥ . . . λm and Let k = max{i | λi ≥ i}. Pm i=1 λi = n. In the Young–diagram of λ, you can see k as follows: it is the size of the largest square you can put into the Young–diagram of λ. (This square is called Durfee– square.) Then one can decompose the Young–diagram into three parts: (1) a square of size k ×k, (2) a partition with at most k parts: λ02 = (λ1 −k, . . . , λk − k), (3) a partition whose parts have size at most k: λ03 = (λk+1 , . . . , λm ). Hence ! ∞ ! ∞ ∞ ∞ X X X X 2 xk p≤k (n2 )xn2 p(n)xn = p≤k (n3 )xn3 = n=0 n2 =0 k=0 = ∞ X k=0 n3 =0 2 xk . ((1 − x)(1 − x2 ) . . . (1 − xk ))2 On the other hand, ∞ X n=0 n p(n)x = ∞ Y 1 . n 1 − x n=1 4 6. First solution: inductive proof. The statement is trivial for n = 0, 1. Assume that the statement holds true till n. Then for n + 1 we have ! n X n (x + y)n+1 = (x + y)n (x + y) = xk y n−k (x + y) = k q k=0 n n X X n n = q n−k xk+1 y n−k + xk y n+1−k = k k q q k=0 k=0 ! n+1 n+1 X X n n n+1 n−k+1 k n+1−k = q + x y = xk y n+1−k . k − 1 k k q q q k=0 k=0 In the last step we used the recursion formula for q-binomial coefficients. Second solution: combinatorial proof. Let us consider a term w in the expansion of (x + y)n which contains k pieces of x’s and (n − k) pieces of y’s, the set of these words will be denoted by Sk . Then w = q A(w) xk y n−k , where A(w) can be computed as follows: for each y we count the number of x’s after y, and we add these numbers together. If we reverse w and replace each x by a letter r ("right") and each y be a letter u ("up"). Then this new word will describe a walk in the (n − k) × k from the bottom-left corner to the up-right corner. Then A(w) counts the number of boxes above the walk, because every time we step upward the number of boxes in the corresponding row is exactly the number of right steps before this up step (so originally these were the number of x’s after this y). Hence n X n X X X q A(w) xk y n−k = w= (x + y)n = k=0 w∈Sk k=0 w∈Sk = n X X q k=0 since we learned that |λ| k n−k x y k=0 λ<(n−k)×k n k q = P 7. We know that λ<(n−k)×k Q λ f = n! = n X n k xk y n−k q q |λ| . i<j ((λi − i) − (λj − Qm i=1 (λi + m − i)! j)) . To prove that f λ = n! det(A), we will use the following fact: let V = V (x1 , . . . xn ) be the m × m matrix such that Vij = xj−1 . This i Q is the so-called Vandermonde determinant, and it is known that det V = i<j (xj − xi ). A simple observation is the following: if P0 , P1 , . . . , Pm−1 are monic polynomials such that 5 deg Pi = i − 1, then for theQ matrix S = S(x1 , . . . xm ) for which Sij = Pj (xi ) we still have det S = det V = i<j (xj − xi ). It is true for the following reason: if P Pi (x) = xi−1 + k<i−1 ak xk , then one can obtain S from V by simple column operations: just add ak times the k + 1.th column to the i.the column. If we multiply the i.th row of A with (λi + m − i)! then we get that 1 det B, i=1 (λi + m − i)! det(A) = Qm where Bij = Pj (xi ), where xi = λi + m − i and P1 (x) = x(x − 1) . . . (x − m + 2), P2 (x) = x(x − 1) . . . (x − m + 3), . . . , Pm (x) = 1. Hence the polynomials Pj are monic polynomials such that deg Pj = m − j. Thus we can apply our previous observation. Note that the order of columns is reversed compared to the Vandermonde-matrix, so Y Y det B = (−1)m(m−1)/2 (xj − xi ) = (xi − xj ) = i<j = i<j Y Y ((λi + m − i) − (λj + m − j)) = ((λi − i) − (λj − j)). i<j i<j Hence Q n! det A = n! i<j ((λi − i) − (λj − Qm i=1 (λi + m − i)! j)) = f λ. 8. First solution: spectral graph theory. Recall that when we determined the eigenvalues of the bipartite graph Gn induced by two neighboring levels, Yn−1 ∪ Yn , of the Young-lattice then we developed a method finding the eigenvalues and eigenvectors of Gn+1 from Gn : if (xn , xn−1 ) is an eigenvector of Gn belonging to λ, then ( √λ12 +1 Un xn , xn ) is an eigenvector of Gn+1 belong√ ing to λ2 + 1. In particulat, this means that the eigenvector corresponding 1 to the largest eigenvalue of Gn+1 is (U n e∅ , √n+1 U n+1 e∅ ) corresponding to the √ eigenvalue n + 1. We have seen at the lecture that X U n e∅ = f λ eλ λ`n and U n+1 e∅ = X µ`n+1 f µ eµ . 6 1 Since (U n e∅ , √n+1 U n+1 e∅ ) corresponding to the eigenvalue X f µ = (n + 1)f λ . √ n + 1 we have µ>λ µ`n+1 Second solution: counting walks. We learned that for a an operator w = A1 A2 . . . Ak , where each Ai are U or D, the number α(w, λ) counting the number of walks of type w starting at ∅ and ending at λ is equal to cw f λ , where Y cw = (bi − ai ), i∈Sw where Sw is the set of indices i for which Ai = D, and bi − ai is the the level where the walk is before we make the step Ai = D. In particular, α(DU n+1 , λ) = (n + 1)f λ . On the other hand, α(DU n+1 , λ) = X α(U n+1 , µ) = µ>λ µ`n+1 X f µ. µ>λ µ`n+1 Hence X f µ = (n + 1)f λ . µ>λ µ`n+1 Third solution: RSK algorithm. Let (F, k) be a pair where F is a standard Young-tableaux and 1 ≤ k ≤ n + 1. Clearly, the number of such pairs is exactly (n + 1)f λ . Let us increase each elements of F by one which are at least k, and then let us insert k into F as we learned at the RSK-algorithm. Then we get a standard Young-tableaux of shape µ Pwhere µµ > λ. Of course, the number of these standard Young-tableauxs is µ>λ f . It is easy to see that µ`n+1 this is a bijection: from the extra square we can carry out the RSK algorithm backwards to learn which k was inserted. 9. We will show that X f (λ)f (λ) = (mn + 1)f m×n , λ where f m×n is the number of standard Young-tableuaxs with shape of the m×n rectangle. Indeed, the right hand side counts the standard Young-tableauxs with shape of the m × n rectangle together with a number k between 0 and 7 nm; this gives us two standard Young-tableauxs, the standard Young-tableaux with boxes containing 1, 2, . . . , k and the standard Young-tableaux with boxes containing k + 1, k + 2, . . . , mn. The left hand side counts it as follows: fix k, take a standard Young-tableaux with shape λ of k boxes, then choose a standard Young-tableaux with shape λ and switch each number i to nm+1−i. Putting together the two standard Young-tableauxs we get a standard Youngtableuax with shape of the m × n rectangle (why?). 10. First solution: combinatorial proof. Let q(n) be the number of partitions whose Young–tableaux are self-conjugate. Then clearly p(n) ≡ q(n) (mod 2) since every other partition can be paired with its conjugate. The surprising fact is that q(n) = r(n) which can be seen as follows. Decompose the self-conjugate Young-tableauxs by central hooks: hooks belonging to some square (i, i). The central hooks have odd lengths and they have different sizes. This gives a partition of n into distinct odd numbers. And of course, the process can be reversed: from a partition of n into distinct odd numbers one can construct a self-conjugate partition. Second solution: generating functions. We count everything mod 2. For instance, ∞ ∞ ∞ Y X Y 1 1 n p(n)x = = n 1−x 1 + xn n=1 n=0 n=1 since −1 = 1 in F2 . Note that 1 1−x since every number can be uniquely written up as a sum of distinct 2-powers. Similarly, for every (odd) m we have ∞ ∞ Y X 1 m2i (1 + x ) = xkm = . 1 − xm i=0 k=0 (1 + x)(1 + x2 )(1 + x4 )(1 + x8 ) · · · = 1 + x + x2 + · · · = Hence ∞ Y ∞ Y Y Y Y 1 1 m (1 − x ) = (1 + xm ). = = i n m2 1+x 1+x n=1 m odd i=0 m odd m odd In the last step we again used that we count in F2 . Since ∞ Y X m (1 + x ) = r(n)xn m odd we are done. n=0 8 11. Note that we only need to determine the eigenvalues of Gn,k for 2k ≤ n since Gn,k is isomorphic to Gn,n−k−1 . For 2k ≤ n we will show by induction that the eigenvalues of Gn,k are the following: let v u k uX aj = t (n − 2i), i=j n then the eigenvalues are ±aj with multiplicity nj − j−1 for j = 0, 1, . . . , k, n n n and 0 with multiplicity k+1 − k . (Note that −1 = 0 which means that the n multiplicity of a0 = 1, and if n = 2k + 1 then k+1 − nk = 0, so 0 is actually not an eigenvalue in this case.) We will completely mimic the proof which we learned when we computed the eigenvalues of the bipartite graphs induced by the Young-lattice. For every set S, let us introduce a vector eS such that all these vectors are linearly independent. Let RPk be the vectorspace induced by the vectors eS , where |S| = k. Let us consider the operators Uk and Dk such that X Uk eS = eT , S⊂T |T |=k+1 and Dk eS = X eT . T ⊂S |T |=k−1 We have seen at the lecture that Dk+1 Uk − Uk−1 Dk = (n − 2k)I. Note that if x = (xk , xk+1 ) is an eigenvector of the adjacency matrix of the graph Gn,k corresponding to some eigenvalue λ, then it is equivalent with saying that Uk xk = λxk+1 and Dk+1 xk+1 = λxk . Dk xk . Then This means that if we introduce the vector xk−1 = √ 2 1 λ −(n−2k) p Dk xk = λ2 − (n − 2k)xk−1 , and 1 1 Uk−1 xk = p Uk−1 Dk xk = p (Dk+1 Uk −(n−2k)I)xk = λ2 − (n − 2k) λ2 − (n − 2k) p 1 =p (λ2 − (n − 2k))xk = λ2 − (n − 2k)xk . λ2 − (n − 2k) 9 This meanspthat (xk−1 , xk ) is an eigenvector of Gn,k−1 corresponding to the eigenvalue λ2 − (n − 2k). This construction works backwards: if (xk−1 , xk ) is an eigenvector of Gn,k−1 corresponding to the eigenvalue µ then (xk , √ 2 1 Uk xk ) is an eigenvector µ +(n−2k) Gn,k . Indeed, if xk+1 =√ 1 Uk xk µ2 +(n−2k) Uk xk = then p µ2 + (n − 2k)xk+1 , and Dk+1 xk+1 = p 1 µ2 1 Dk+1 Uk xk = p (Uk−1 Dk +(n−2k)I)xk = 2 + (n − 2k) µ + (n − 2k) p 1 =p (µ2 + (n − 2k))xk = µ2 + (n − 2k)xk . µ2 + (n − 2k) Note that this backward direction means that if µ is an eigenvalue of Gn,k−1 p 2 with multiplicity mµ , then µ + (n − 2k) is an eigenvalue of Gn,k with multiplicity at least mµ . (As we will see, it is enough to consider only this backward direction, but the other direction help us to see that actually there is a bijection between the positive eigenvalues and eigenspaces of Gn,k and the non-negative eigenvalues and eigenspaces of Gn,k−1 .) Note that for any bipartite graph G with classes A and B for which |B| ≥ |A|, the rank of the adjacency matrix of G is at most 2|A|, consequently, the multiplicity of 0 is at least |A| + |B| − 2|A| = |B| − |A|. Now we are ready √to prove our claim. Gn,0 is a star on n + 1 vertices so its eigenvalues are ± n with multiplicity 1, and 0 with multiplicity n − 1 = n n − . So the statement is true for k = 0. Assume that the statement holds 1 0 true for k − 1. Then the eigenvalues of Gn,k−1 are v u k−1 uX ±t (n − 2i) i=j n n with multiplicity j − j−1 for j = 0, 1, . . . , k − 1, and 0 with multiplicity n n − k−1 . Then the eigenvalues of Gn,k are k v u k uX ±t (n − 2i) i=j 10 n with multiplicity at least nj − j−1 for j = 0, 1, . . . , k, and 0 with multiplicity n n at least k+1 − k . But since k X n n n n n n − +2 − = + = |V (Gn,k )|, k+1 k j j−1 k+1 k j=0 we have found all eigenvalues and multiplicities: v u k uX ±t (n − 2i) i=j with multiplicity exactly n exactly k+1 − nk . n j − n j−1 for j = 0, 1, . . . , k, and 0 with multiplicity 12. Let m = α1 + · · · + αr , and let G = Sα1 × · · · × Sαr ≤ Sm embedded into Sm such that Sα1 acts on {1, . . . , α1 }, Sα2 acts on {α1 +, . . . , α1 + α2 } etc. Then the poset Sm /G is isomorphic to the poset of divisors of n (why?). Hence we can apply all the theorems we learned: the poset is rank-symmetric, unimodal, Sperner. This problem has an elementary solution going as follows: if you have two posets with one-one symmetric chain decompositions then you can create a symmetric chain decomposition to the product poset. Note that all you have to do is to create a symmetric chain decomposition of the products of two chains and it is quite easy.