跳到主要内容

東京大学 情報理工学系研究科 コンピュータ科学専攻 2015年8月実施 専門科目I 問題1

Author

祭音Myyura (co-authored with GPT 5.6 SOL)

Description

Given two nn-dimensional integer vectors xx and yy, let us write xyx\doteq y if we have xiyi(mod2)x_i\equiv y_i\pmod 2 for each i[1,n]i\in[1,n]. Here xix_i and yiy_i denote the ii-th elements of the vectors xx and yy, respectively. Likewise, given two n×nn\times n integer matrices AA and BB, we write ABA\doteq B if we have aijbij(mod2)a_{ij}\equiv b_{ij}\pmod 2 for each i,j[1,n]i,j\in[1,n]. Here aija_{ij} and bijb_{ij} denote the elements in the ii-th row and the jj-th column of the matrices AA and BB, respectively.

In what follows, a vector all of whose elements are either 00 or 11 is referred to as a “0-1 vector”. A vector each of whose elements is chosen from 00 and 11, with equal probabilities and independently from the other elements, is referred to as a “random 0-1 vector”. The zero vector (i.e., the vector all of whose elements are 00) is denoted by oo; and the zero matrix (i.e., the matrix all of whose elements are 00) is denoted by OO.

Answer the following questions.

(1) Let x{0,1}3x\in\{0,1\}^3 be a random 0-1 vector. Derive the probability with which

(011101110)xo\begin{pmatrix} 0&1&1\\ 1&0&1\\ 1&1&0 \end{pmatrix}\cdot x\doteq o

holds.

(2) Let AA be an n×nn\times n integer matrix that does not satisfy AOA\doteq O, and x{0,1}nx\in\{0,1\}^n be a random 0-1 vector. Prove that the probability with which AxoA\cdot x\doteq o holds is no greater than 1/21/2.

(3) Let AA, BB and CC be n×nn\times n matrices that do not satisfy ABCA\cdot B\doteq C, and x{0,1}nx\in\{0,1\}^n be a random 0-1 vector. Prove that the probability with which ABxCxA\cdot B\cdot x\doteq C\cdot x holds is no greater than 1/21/2.

(4) Show an O(n2)O(n^2) algorithm that: takes three n×nn\times n integer matrices AA, BB and CC; always answers “SATISFIED” if the condition ABCA\cdot B\doteq C is satisfied; and answers “NOT SATISFIED”, with a probability greater than 9/109/10, if the condition ABCA\cdot B\doteq C is not satisfied.

题目描述

向量、矩阵之间的等号均按元素模 22 理解。随机 00-11 向量是指各分量独立且等概率取 0,10,1 的向量,零向量和零矩阵分别记为 o,Oo,O。回答下列问题。

(1)对随机向量 x{0,1}3x\in\{0,1\}^3,求

(011101110)x=o\begin{pmatrix} 0&1&1\\ 1&0&1\\ 1&1&0 \end{pmatrix}x=o

成立的概率。

(2)设 AA 是不满足 A=OA=On×nn\times n 整数矩阵,x{0,1}nx\in\{0,1\}^n 为随机向量。证明 Ax=oAx=o 的概率不超过 1/21/2

(3)设 A,B,CA,B,C 是不满足 AB=CAB=Cn×nn\times n 整数矩阵,x{0,1}nx\in\{0,1\}^n 为随机向量。证明 ABx=CxABx=Cx 的概率不超过 1/21/2

(4)给出一个 O(n2)O(n^2) 随机算法:若 AB=CAB=C,总回答“成立”;若 ABCAB\ne C,则以严格大于 9/109/10 的概率回答“不成立”。

Kai

以下计算均在域 F2\mathbb F_2 上进行。

(1)

三个方程分别给出 x2=x3x_2=x_3x1=x3x_1=x_3x1=x2x_1=x_2,故只有 x=(0,0,0)T,(1,1,1)Tx=(0,0,0)^\mathsf T,(1,1,1)^\mathsf T 两个解。因此

Pr(Ax=o)=223=14.\Pr(Ax=o)=\frac{2}{2^3}=\boxed{\frac14}.

(2)

AA 的一个非零行,并在该行中取系数为 11 的位置 jj。固定除 xjx_j 外的所有分量后,该行与 xx 的内积形如

xj+c(mod2).x_j+c\pmod 2.

由于 xjx_j 等概率取 0,10,1,该内积为 00 的条件概率恰为 1/21/2。事件 Ax=oAx=o 还要求其余各行也为 00,故

Pr(Ax=o)12.\boxed{\Pr(Ax=o)\le \frac12}.

(3)

D=ABCD=AB-C。由 ABCAB\ne CDOD\ne O,且

ABx=Cx    Dx=o.ABx=Cx\iff Dx=o.

由(2)立即得到所求概率不超过 1/21/2

(4)

独立重复以下检验 44 次:随机生成 x{0,1}nx\in\{0,1\}^n,依次计算

y=Bx,z=Ay,t=Cx(mod2).y=Bx,\qquad z=Ay,\qquad t=Cx\pmod2.

若某次 ztz\ne t,回答“不成立”;四次均相等时回答“成立”。每次只做三次矩阵向量乘法,故总时间为 O(n2)O(n^2)

AB=CAB=C,算法必定回答“成立”。若 ABCAB\ne C,由(3),四次均未检出错误的概率至多为 24=1/162^{-4}=1/16,所以回答“不成立”的概率至少为

1116=1516>910.1-\frac1{16}=\frac{15}{16}>\frac9{10}.