Chapter 7

Algebraic Methods

[20/45%]
Show
LaTeX
Problem 7.1.1

Recall that for a set FF one defines Fn:={(a1,a2,…,an):a1,a2,…,an∈F}F^{n} := \left\{ (a_{1}, a_{2}, \ldots , a_{n}) : a_{1}, a_{2}, \ldots , a_{n} \in F\right\}. The elements of this set are called vectors (or tuples, or points). If F∈{Z2,Z,Q,R}F \in \left\{ \mathbb {Z}_{2}, \mathbb {Z}, \mathbb {Q}, \mathbb {R}\right\}, then vectors can be added componentwise: (x1,…,xn)+(y1,…,yn):=(x1+y1,…,xn+yn)(x_{1}, \ldots , x_{n}) + (y_{1}, \ldots , y_{n}) := (x_{1}+y_{1}, \ldots , x_{n}+y_{n}). If F∈{Z,Q,R}F \in \left\{ \mathbb {Z}, \mathbb {Q}, \mathbb {R}\right\}, then a vector can be multiplied componentwise by a number λ∈F\lambda \in F: λ(x1,…,xn):=(λx1,…,λxn)\lambda (x_{1}, \ldots , x_{n}) := (\lambda x_{1}, \ldots , \lambda x_{n}) (this can also be done for F=Z2F = \mathbb {Z}_{2}, but it is not interesting).

  1. Among any n+1n+1 tuples of length nn consisting of zeros and ones, there are several (a nonzero number of) tuples whose componentwise sum modulo two is the zero tuple.

  2. For any n+1n+1 vectors v1,…,vn+1∈Qnv_{1}, \ldots , v_{n+1} \in \mathbb {Q}^{n} there exist rational numbers λ1,…,λn+1\lambda_{1}, \ldots , \lambda_{n+1}, not all equal to zero, for which λ1v1+…+λn+1vn+1=(0,…,0)\lambda_{1} v_{1} + \ldots + \lambda_{n+1} v_{n+1} = (0, \ldots , 0).

  3. The analog of theorem (Q)(\mathbb {Q}) holds for the real numbers, the complex numbers, and the integers.

The tuples from Problems 7.1.1(Z2)(\mathbb {Z}_{2}), (Q)(\mathbb {Q}) are called linearly dependent — over Z2\mathbb {Z}_{2} and over Q\mathbb {Q}, respectively. Linear independence is the negation of linear dependence. Linear (in)dependence of polynomials over Z2\mathbb {Z}_{2} and Q\mathbb {Q}, respectively, is defined similarly.

?
Remark.
?

Remark. Similar definitions can be given in a more general situation — this leads to the notions of a ring and of a module over it. An attempt to prove (and use!) an analog of the linear dependence theorem (Problems 7.1.1(Z2)(\mathbb {Z}_{2}), (Q)(\mathbb {Q})) leads to the notions of a field and of a linear space over it. (In the case of the integers, not all generalizations go through anymore.) Details can be found in a textbook on linear algebra.

The phrase $N$ elements'' (in particular, subsets) means NN pairwise distinct elements'' (in particular, subsets).

Problem 7.1.2

A linear subspace is a subset L⊂QnL \subset \mathbb {Q}^{n} closed under vector addition and multiplication by rational numbers. A linear subspace LL is called nn-dimensional if there exist linearly independent vectors v1,…,vn∈Lv_{1}, \ldots , v_{n} \in L such that every vector v∈Lv \in L is a linear combination of these vectors, i.e. there exist numbers λ1,…,λn∈Q\lambda_{1}, \ldots , \lambda_{n} \in \mathbb {Q} for which v=λ1v1+…+λnvnv = \lambda_{1} v_{1} + \ldots + \lambda_{n} v_{n}. The number nn is called the dimension of the space LL.

Let F\mathscr {F} be a family of subsets of the set Rn\mathscr {R}_{n}.

?
(1)

If every subset in F\mathscr {F} has an odd number of elements and the intersection of any two subsets in F\mathscr {F} has an even number of elements, then ∣F∣≤n\left|\mathscr {F}\right| \leq n.

(2)

Construct an example where this bound is attained.

(3)

If the intersection of any two subsets in F\mathscr {F} has exactly qq elements and every subset in F\mathscr {F} has more than qq elements, then ∣F∣≤n\left|\mathscr {F}\right| \leq n.

(4)

If q>0q > 0 and the intersection of any two subsets in F\mathscr {F} has exactly qq elements, then ∣F∣≤n\left|\mathscr {F}\right| \leq n.

Problem 7.1.3
?
(1)

There exist 2k2^{k} subsets of a 2k2k-element set such that each of them has an even number of elements and the intersection of any two of them has an even number of elements.

(2)

There cannot be more than 2k2^{k} subsets satisfying the conditions of part (1).

Problem 7.1.4

The distance between points of the space Rn\mathbb {R}^{n} is defined by the formula

∣(x1,…,xn),(y1,…,yn)∣:=(x1−y1)2+…+(xn−yn)2. \left|(x_{1}, \ldots , x_{n}), (y_{1}, \ldots , y_{n})\right| := \sqrt{(x_{1}-y_{1})^{2} + \ldots + (x_{n}-y_{n})^{2}}.
?
(1)

The largest number of points in Rn\mathbb {R}^{n} with equal pairwise distances is n+1n+1.

(2)

Construct n(n−1)2\dfrac {n(n-1)}{2} points in Rn\mathbb {R}^{n} whose pairwise distances take only two distinct values.

(3)

For a∈Ra \in \mathbb {R} and points v=(v1,…,vn)∈Rnv = (v_{1}, \ldots , v_{n}) \in \mathbb {R}^{n}, x=(x1,…,xn)∈Rnx = (x_{1}, \ldots , x_{n}) \in \mathbb {R}^{n}, denote Pv(x):=∣x−v∣2−a2P_{v}(x) := \left|x-v\right|^{2} - a^{2}. If the pairwise distances between kk points u1,…,uk∈Rnu_{1}, \ldots , u_{k} \in \mathbb {R}^{n} are equal to aa, then the polynomials Pu1,…,PukP_{u_{1}}, \ldots , P_{u_{k}} are linearly independent over Q\mathbb {Q}.

(4)

If the pairwise distances between kk points in Rn\mathbb {R}^{n} take only two distinct values, then k≤(n+1)(n+4)2k \leq \dfrac {(n+1)(n+4)}{2}.

Problem 7.1.5

For n,k∈Zn, k \in \mathbb {Z}, denote

Vn,k:={(x1,…,xn)∈{0,1}n:∑sxs=k}. V_{n,k} := \left\{ (x_{1}, \ldots , x_{n}) \in \left\{ 0,1\right\} ^{n} : \sum _{s} x_{s} = k\right\} .
?
(1)

Among any 327327 pairwise intersecting 99-element subsets of a 2525-element set, there are two subsets whose intersection has exactly 33 or exactly 66 elements.

(2)

Among any 327327 points in V25,9V_{25,9}, there are two whose scalar product lies in {0,3,6}\left\{ 0,3,6\right\}.

(3)

For any a→∈V25,9\overrightarrow {a} \in V_{25,9}, expand the product

(a→⋅(x1,x2,…,x25)−1)(a→⋅(x1,x2,…,x25)−2), (\overrightarrow {a} \cdot (x_{1}, x_{2}, \ldots , x_{25}) - 1)(\overrightarrow {a} \cdot (x_{1}, x_{2}, \ldots , x_{25}) - 2),

where x1,x2,…,x25x_{1}, x_{2}, \ldots , x_{25} are variables. In each of the resulting monomials, perform the following operation: for each ii, if the monomial contains the factor xi2x_{i}^{2}, replace this factor with 11. Denote the resulting polynomial by Fa→(x1,…,x25)F_{\overrightarrow {a}}(x_{1}, \ldots , x_{25}). Prove that if the scalar product of no two vectors among a→1,…,a→s∈V25,9\overrightarrow {a}_{1}, \ldots , \overrightarrow {a}_{s} \in V_{25,9} is divisible by 33, then the polynomials Fa→1,…,Fa→sF_{\overrightarrow {a}_{1}}, \ldots , F_{\overrightarrow {a}_{s}} are linearly independent over Q\mathbb {Q}.

(4)

Specify 326326 polynomials whose linear combinations with rational coefficients yield every polynomial Fa→F_{\overrightarrow {a}}, a→∈V25,9\overrightarrow {a} \in V_{25,9}.

Problem 7.1.6
?
(1)

Among any 107107 five-element subsets of a 1414-element set, there are two subsets whose intersection has exactly 22 elements.

(2)

The same for 9393 subsets.

(3)

The same for 9292 subsets.

(4)

It is impossible to color all five-element subsets of a 1414-element set with 2121 colors so that any two five-element subsets intersecting in exactly two elements have different colors.

Remark.
?

Cf.the remark in Problem 5.1.4. Here is an equivalent formulation. The vertices of the graph are all five-element subsets of a 1414-element set. Its edges are the pairs of subsets intersecting in exactly two elements. Prove that this graph cannot be properly colored with 2121 colors.

Problem 7.1.7
?
(1)

For a prime pp and an integer tt, the number G(t):=(t−1)(t−2)…(t−p+1)G(t) := (t-1)(t-2)\ldots (t-p+1) is divisible by pp if and only if tt is not divisible by pp.

(2)

Let pp be a prime and n=4pn = 4p. Denote

M={(1,y2,y3,…,yn):yk∈{1,−1} and among y2,…,yn the number of minus ones is even}. M = \left\{ (1, y_{2}, y_{3}, \ldots , y_{n}) : y_{k} \in \left\{ 1,-1\right\} \text{ and among } y_{2}, \ldots , y_{n} \text{ the number of minus ones is even}\right\} .

Denote G(t):=(t−1)(t−2)…(t−p+1)G(t) := (t-1)(t-2)\ldots (t-p+1). For any a→∈M\overrightarrow {a} \in M, expand the brackets in the product G(a→⋅(1,x2,…,xn))G(\overrightarrow {a} \cdot (1, x_{2}, \ldots , x_{n})), where x2,…,xnx_{2}, \ldots , x_{n} are variables. In each of the resulting monomials, for each ii we replace xi2x_{i}^{2} by 11 as long as possible. Denote the resulting polynomial by Fa→(x2,…,xn)F_{\overrightarrow {a}}(x_{2}, \ldots , x_{n}).

Prove that if no two vectors among a→1,…,a→s∈M\overrightarrow {a}_{1}, \ldots , \overrightarrow {a}_{s} \in M have zero scalar product, then the polynomials Fa→1,…,Fa→sF_{\overrightarrow {a}_{1}}, \ldots , F_{\overrightarrow {a}_{s}} are linearly independent over Q\mathbb {Q}.

(3)

There exist nn and a bounded subset of Rn\mathbb {R}^{n} that cannot be partitioned into n+1n+1 nonempty parts of smaller diameter.

Problem 7.1.8
?
(1)

If tt is prime, then among any 1+∑j=0t−1(nj)1 + \sum \limits_{j=0}^{t-1} \dbinom {n}{j} distinct subsets of an nn-element set, each of which has kk elements, there are two subsets whose intersection has a number of elements divisible by tt.

(2)

The same, except that an integer qq is given and is divisible by $t$'' is replaced by is congruent to qq modulo tt''.

Problem 7.1.9
?
(1)

If the edge set of the graph KnK_{n} is the union of the edge sets of ss edge-disjoint complete bipartite graphs, then s≥n−1s \geq n-1.

(2)

Construct a family of bipartite graphs for which this bound is attained.

Remark.
?

A more detailed and extended exposition can be found in [Mk, R1], and a more advanced one — in [BF].

Problem 7.2.1

If a matrix AA of size n×nn \times n has all entries of absolute value at most 11, then ∣det⁡(A)∣≤nn/2\left|\operatorname {det}\left(A\right)\right| \leq n^{n/2}.

A square matrix HH is called a Hadamard matrix if all its entries are equal to ±1\pm 1 and H⋅HT=nEnH \cdot H^{T} = nE_{n}, where nn is the order of the matrix HH and EnE_{n} is the identity matrix.

?
Problem 7.2.2

Construct an n×nn \times n Hadamard matrix for

?
(1)

n=2n = 2;

(2)

n=4n = 4;

(3)

n=8n = 8;

(4)

n=16n = 16;

(5)

n=12n = 12.

Problem 7.2.3
?
(1)

Any two columns of a Hadamard matrix are orthogonal.

(2)

A matrix is a Hadamard matrix if and only if its entries are equal to ±1\pm 1 and any two of its rows are orthogonal.

(3)

Hadamard matrices attain the upper bound in Hadamard's theorem 7.2.1. (Hadamard matrices got their name owing to this result.)

(4)

If there exists an n×nn \times n Hadamard matrix and n>2n > 2, then nn is divisible by 44.

Remark.
?

Conjecture. An n×nn \times n Hadamard matrix exists for every number nn divisible by 44.

The conjecture has not been proved even for some numbers less than 10001000, namely: for 668,716,892668, 716, 892.

Problem 7.2.4

Solving this problem and the next one (7.2.5) requires the basic properties of quadratic residues; see [GIM, § 9], [Vi, § 5], [ZSS, Sec. 3.3].

For a prime number pp, denote Sd=Sp,d:=∑j∈Zp(j(j+d)p)S_{d} = S_{p,d} := \sum \limits_{j \in \mathbb {Z}_{p}} \left(\dfrac {j(j+d)}{p}\right) (this is a sum of Legendre symbols).

?
(1)

Prove that SdS_{d} does not depend on d≠0d \neq 0.

(2)

Find SdS_{d} for each d∈Zpd \in \mathbb {Z}_{p}.

Problem 7.2.5

Solving this problem (along with the previous one, 7.2.4) requires the basic properties of quadratic residues; see [GIM, § 9], [Vi, § 5], [ZSS, Sec. 3.3].

Construct an n×nn \times n Hadamard matrix for

?
(1)

n=2an = 2a if there exists an a×aa \times a Hadamard matrix;

(2)

n=abn = ab if there exist a×aa \times a and b×bb \times b Hadamard matrices;

(3)

n=p+1n = p+1, where pp is a prime number of the form 4k−14k-1;

(4)

n=2p+2n = 2p+2, where pp is a prime number of the form 4k+14k+1.

Problem 7.2.6

A Hadamard matrix is called normalized if its first row and first column consist entirely of ones.

Write out all normalized Hadamard matrices of orders 1;2;41; 2; 4.

?
Problem 7.2.7

The Hadamard property of a matrix is preserved under the following transformations:

?
(1)

multiplying a row or a column by −1-1;

(2)

interchanging rows or columns.

Hadamard matrices obtained from each other by applying some number of transformations (1) and (2) are called equivalent.

Problem 7.2.8
?
(1)

Which of the matrices from Problem 7.2.6 are equivalent?

(2)

Every Hadamard matrix is equivalent to some normalized one.

The number of equivalence classes: for orders 1,2,4,8,121, 2, 4, 8, 12 — 11, 1616 — 55, 2020 — 33, 2424 — 6060, 2828 — 487487, 3232 — more than a million.

Problem 7.2.9

Is it true that for any Hadamard matrices HH and H′H' the matrices H⊗H′H \otimes H' and H′⊗HH' \otimes H are equivalent? Here the tensor product ⊗\otimes is defined in the hint to Problem 7.2.5 (1, 2).

?
Problem 7.2.10

The Hadamard matrix HH built by means of the (Paley) construction from Problem 7.2.5 (4) is equivalent to the matrix HTH^{T}.

?
Problem 7.2.11

Does there exist a Hadamard matrix HH that is not equivalent to the matrix HTH^{T}?

?