115 年 國立臺灣大學資料科學碩士學位學程《工程數學(D)》
第 1 題2 分
- (2%) Let be a vector space, which of the following is not a consequence?
(a) If , then .
(b) If , then .
(c) If , then .
(d) If , then .
登入後即可作答並保存紀錄。
核心觀念
向量空間 必須滿足:
- 對向量加法封閉,且每個向量都有加法反元素;
- 純量乘法定義於純量與向量之間;
- 向量加法具有交換律與結合律。
題目要求判斷哪一項不是向量空間公理或其直接推論。
解題方法
逐一檢查各選項是否能由向量空間的基本性質推出。
選項分析
(a) 若 ,則 。
由向量空間對加法封閉,且 ,可得
因此此敘述正確,是向量空間性質的直接推論。
(b) 若 ,則 。
純量 可寫成 ,利用純量乘法對純量加法的分配律:
因此此敘述正確。
(c) 若 ,則 。
一般向量空間只定義:
第 2 題2 分
- (2%) Which of the following is not able to serve as a vector space over the field of real numbers ?
(a) The complex numbers .
(b) The set of all real numbers such that .
(c) The functions that are differentiable.
(d) The lattice where is an integer greater than 2026.
(e) The set where we use as and as .
登入後即可作答並保存紀錄。
核心觀念
判斷一個集合是否為實數域 上的向量空間,必須檢查:
- 向量加法下封閉;
- 實數純量乘法下封閉;
- 含有零向量;
- 每個向量都有加法反元素;
- 加法與純量乘法滿足向量空間公理。
本題的關鍵是:即使某集合具有加法結構,只要對任意實數純量 不封閉,就不是 上的向量空間。
解題方法
逐一檢查各選項是否能定義成 上的向量空間,特別注意純量乘法是否對所有 都成立。
選項分析
(a) 複數集合
將複數視為向量,定義通常的複數加法與實數純量乘法:
若 ,則 ;若 ,則 。零向量為 , 的加法反元素為 。
事實上, 可寫成
因此它是 上的二維向量空間,基底可取 。
所以 (a) 可以作為 上的向量空間。
(b) 滿足 的所有實數
對任意實數 ,皆有
因此 只有在 時成立,集合為
單元素集合 是平凡向量空間:
- 零向量是 ;
- ,加法封閉;
- 對任意 ,,純量乘法封閉。
所以 (b) 可以作為 上的向量空間。
(c) 所有可微函數
令 為可微函數,。定義逐點運算:
因為可微函數的和仍可微,且常數倍仍可微:
零向量是零函數
而 的加法反元素為 ,其中
因此所有 的可微函數構成 上的向量空間。
第 3 題2 分
- (2%) According to the most strict definition of linear transformation, which of the following is not a linear transformation?
(a) .
(b)
(c) The sum of another two linear transformations.
(d) Mapping any vector to the real number .
登入後即可作答並保存紀錄。
線性變換 必須滿足
(a)
兩條性質皆成立 → 線性。
(b)
此為恆等映射 ,顯然滿足加法與純量乘法 → 線性。
(c) 兩個線性變換 的和
第 4 題2 分
- (2%) Which of the following can be used as the definition of the rank of a matrix M?
(a) The width of M if M is invertible.
(b) The dimension of the right null space .
(c) The number of basis vectors of the column space of M.
(d) The number of 1's in the reduced row echelon form of M.
登入後即可作答並保存紀錄。
核心觀念
矩陣 的 rank(秩)定義為其 column space 的維度:
因此,rank 也等於:
- column space 中基底向量的數目;
- reduced row echelon form(RREF)中的 pivot 數目;
- RREF 中 leading 的數目;
- 線性獨立行或列的最大數目。
另外,右零空間的維度是 nullity(零度),由 rank-nullity theorem 可得:
解題方法
逐一檢查各選項是否符合 rank 的一般定義。特別注意:RREF 中應計算的是 pivot 的數目,而不是所有數值等於 的元素。
選項分析
(a) The width of if is invertible.
錯誤。
若 為可逆矩陣,則 必為 方陣,且
此時矩陣的 width(欄數)確實等於 rank。然而,這只適用於可逆方陣,不能用來定義一般矩陣的 rank。不可逆矩陣或非方陣的 rank 並不等於其 width。
(b) The dimension of the right null space .
錯誤。
集合
第 5 題2 分
- (2%) Professor Gilbert Strang wrote the textbook Introduction to Linear Algebra for undergraduate students. The book is almost 600 pages long. Which of the following topics is on page 58?
(a) Pseudo inverse.
(b) Matrix multiplication.
(c) Positive semi-definite.
(d) Singular value decomposition.
登入後即可作答並保存紀錄。
核心觀念
本題考查對 Gilbert Strang《Introduction to Linear Algebra》內容編排的熟悉程度,並非要求實際計算矩陣。該書約 600 頁,前半部先介紹向量、線性方程組與矩陣基本運算,再進入正定矩陣、廣義逆矩陣與奇異值分解等進階主題。
第 58 頁位於矩陣基本運算的早期章節,主題是 矩陣乘法(Matrix multiplication)。
矩陣乘法的定義為:若
則乘積 存在,且
其第 個元素為
矩陣乘法的核心條件是:前一個矩陣的欄數必須等於後一個矩陣的列數。
解題方法
依照該書的內容順序判斷:
- 矩陣乘法屬於線性代數的基礎運算,通常在全書前段介紹。
- Pseudo inverse、positive semi-definite、singular value decomposition 都屬於後續的進階矩陣主題。
- 因此第 58 頁所對應的主題應是矩陣乘法。
選項分析
(a) Pseudo inverse
Pseudo inverse,即「廣義逆矩陣」,通常記為 ,用於矩陣不可逆或非方陣的情形,並與最小平方解及奇異值分解密切相關。這是較後段的進階主題,不是第 58 頁的內容。
第 6 題2 分
- (2%) On page 187 of Introduction to Linear Algebra, we can see a picture very similar to the following that explains the four fundamental subspaces:
🖼️【此處有附圖,請對照原卷】
(a) row space
(b) null space
(c) column space
(d) rank space
Which of the (a), (b), (c), and (d) is labeled incorrectly?
登入後即可作答並保存紀錄。
核心觀念
矩陣 的四個基本子空間為:
- 列空間(row space)
- 零空間(null space)
- 行空間(column space)
- 左零空間(left null space)
其中「rank space」不是線性代數中四個基本子空間的正式名稱。矩陣的 rank 是秩,代表列空間或行空間的維度,並不是一個獨立的基本子空間。
選項分析
-
(a) row space:正確。
列空間由矩陣 的所有列向量所張成,是四個基本子空間之一。 -
(b) null space:正確。
零空間定義為
也是四個基本子空間之一。
第 7 題2 分
- (2%) Choose one incorrect statement regarding the augmented matrix .
(a) is a column vector.
(b) is a matrix that is at least 2 columns wide.
(c) It can be used to determine if the system has a solution.
(d) It can be used to determine if the system has more than one solutions.
登入後即可作答並保存紀錄。
核心觀念
-
增廣矩陣(Augmented Matrix)的定義
對於包含 個方程式、 個未知數的線性方程組 :- 係數矩陣 的維度為 (其中 )。
- 未知數向量 為 的行向量(column vector)。
- 常數向量 為 的行向量(column vector)。
- 增廣矩陣表示為 ,其維度為 ,是將向量 作為最後一行(column)附加於矩陣 右側所構成。
-
線性系統解的判定(Rouché–Capelli 定理)
透過對增廣矩陣 進行基本列運算(Elementary Row Operations)化簡為列梯形矩陣(Row Echelon Form),可比較 與 來判定解的結構:- 無解(Inconsistent):(增廣矩陣最後一行的常數行出現軸元 pivot)。
- 恰有一解(Unique Solution):(無自由變數)。
- 無限多解(Infinitely Many Solutions):(存在 個自由變數)。
解題方法
本題要求選出關於增廣矩陣 的「錯誤敘述」。
解題切入點在於:
- 檢驗矩陣與向量的基本維度定義(矩陣行數 與向量 的形式)。
第 8 題3 分
- (3%) How many matrices are made of the numbers 0 and 1?
(a) About 10.
(b) About 20.
(c) About 50.
(d) About 100.
(e) About 200.
(f) About 500.
(g) About 1000.
登入後即可作答並保存紀錄。
核心觀念
矩陣共有
個位置。每個位置都可以獨立填入 或 ,因此每個位置有 種選擇。
根據乘法原理,矩陣總數為
解題方法
將矩陣視為 個獨立位置:
每個 有 種填法,因此
約等於 ,所以選擇 (f)。
選項分析
第 9 題3 分
- (3%) What is the determinant of the following matrix?
(a) About .
(b) About .
(c) About .
(d) About .
(e) About .
(f) About .
(g) About .
登入後即可作答並保存紀錄。
核心觀念
本題考查行列式的拉普拉斯展開(按某一行或某一列展開):
其中 是刪去第 列、第 欄後所得的子矩陣行列式。
題目中的第 欄只有一個非零元素,第 欄在刪除後也只有一個非零元素,因此適合連續使用降階展開。
解題方法
原矩陣記為 :
沿第 欄展開。第 欄只有第 列的元素 非零,因此
因為 ,所以
對上述 行列式沿第 欄展開。第 欄只有第 列的元素 非零:
由於 ,得
接著沿第 欄展開,只有第 列的元素 非零:
因此
計算 行列式:
所以
第 10 題3 分
- (3%) Let be a square matrix. Let be the entry on the th row and th column of . Let . Let be the Gershgorin disk . The Gershgorin circle theorem states that every eigenvalue of lies inside at least one of the Gershgorin disks. Consider the matrix
Accordingly, which of the following is not an eigenvalue of the matrix?
(a) -40.0001427912351045
(b) -11.1008990913169226
(c) 21.1196676389242784
(d) 28.9813742436277487
登入後即可作答並保存紀錄。
核心觀念
本題考查 Gershgorin 圓盤定理。若以第 欄計算,
則第 個 Gershgorin 圓盤為
Gershgorin 圓盤定理指出:矩陣的每一個特徵值,至少會落在一個 Gershgorin 圓盤內。因此,若某個數值落在所有圓盤之外,便確定不是特徵值。
解題方法
矩陣為
依題目定義計算各欄的 :
所以四個 Gershgorin 圓盤為
由於所有選項皆為實數,可直接比較其與各圓心的距離:
- 覆蓋實軸區間 ;
- 覆蓋實軸區間 ;
- 覆蓋實軸區間 ;
- 覆蓋實軸區間 。
選項分析
(a)
其與 的距離為
因此落在 內,符合特徵值的位置。
(a) 正確為特徵值。
(b)
其與 的距離為
第 11 題3 分
- (3%) The power method refers to the following iterative process to find the dominant eigenvalue of a matrix T: Let be an initial guess. Then repeat the following steps multiple times:
• ,
• , where is the first component of .
Eventually, will be very close to the dominant eigenvalue of . Now consider the matrix
It is known that the dominant eigenvalue of is . What is the missing digit?
(a) 0
(b) 1
(c) 2
(d) 3
(e) 4
(f) 5
(g) 6
(h) 7 (i) 8 (j) 9
For this problem, we recommend using as the initial guess and repeat the steps 10 times.
登入後即可作答並保存紀錄。
解題重點
- 令 為列向量,依題目迭代
其中 為 的第一個分量。
2. 若 的支配特徵值為 ,則在每一步正規化後, 會收斂至 。
3. 矩陣
因此只要找出 的最大實根 ,則
- 計算 的特徵多項式(展開 )得到
以牛頓法求根,收斂至
因而
第 12 題3 分
- (3%) Jensen is playing with orthogonal matrices, that is, matrices such that . He wants to characterize all real matrices such that, for a very small real number , the matrix is still orthogonal up to small error terms. More precisely, he wants that
is a matrix whose entries are all multiples of with no or terms. Which of the following conditions should satisfy?
(a) is orthogonal.
(b) is orthonormal.
(c) is vertible.
(d) is invertible.
(e) is symmetric.
(f) is skew-symmetric.
登入後即可作答並保存紀錄。
核心觀念
正交矩陣 滿足
題目考察的是正交矩陣的「一階擾動」。若矩陣在微小變化下仍維持正交性至一階,則一階變化量必須滿足
這正是 為斜對稱矩陣(skew-symmetric matrix)的條件。
解題方法
將題目中的矩陣因式分解:
因此
由於 ,可得
所以
題目要求沒有 項,因此必須有
等價於
故 必須是斜對稱矩陣。
此時
其每個元素確實都是 的倍數,符合題目要求。
選項分析
(a) is orthogonal
錯誤。正交矩陣滿足 ,但題目要求的是
兩者是不同條件。例如
第 13 題3 分
- (3%) Lights Out is an electronic game where your goal is to turn off all the lights by pressing switches. In the following diagram, switches are represented by rectangles and lights are represented by circles. Pressing a switch toggles the on/off state of the lights connected to the switch.
🖼️【此處有附圖,請對照原卷】
(a)
(b)
(c)
(d)
(e)
XXIXXI
(off) On) off (off) (On
Which set of switches turn off all the lights if each is pressed once?
登入後即可作答並保存紀錄。
1. 初始狀態與連接關係
根據圖中線路,各開關控制的燈泡集合如下:
- 開關 控制:
- 開關 控制:
- 開關 控制:
- 開關 控制:
- 開關 控制:
定義 表示開關 是否被按下( 表示按下, 表示不按下)。
燈泡的初始狀態為:
2. 建立二元一次方程組
若要使所有燈泡最終皆為關閉狀態(),在模 2 餘數系統()下須滿足:
第 14 題3 分
- (3%) A binary linear block code is generated by the following matrix
The code is said to be maximum distance separable (MDS) if all minors of are odd numbers. However, is not MDS. Which three columns of give rise to an even minor?
(a) {1,2,3} (b) {1,2,4} (c) {1,2,5} (d) {1,2,6} (e) {1,3,4} (f) {1,3,5} (g) {1,3,6} (h) {1,4,5} (i) {1,4,6} (j) {1,5,6} (k) {2,3,4} (l) {2,3,5} (m) {2,3,6} (n) {2,4,5} (o) {2,4,6} (p) {2,5,6} (q) {3,4,5} (r) {3,4,6} (s) {3,5,6} (t) {4,5,6}
登入後即可作答並保存紀錄。
核心觀念
對二元線性區塊碼而言,生成矩陣 的所有 子矩陣,其行列式若為奇數,等價於在模 下行列式為 。題目所稱的 MDS 條件為:
其中 表示取第 欄所形成的 矩陣。
因此,只要三個欄向量在 上線性相關,其行列式即為偶數,該三欄便會產生偶數 minor。
將各欄記為
解題方法
直接觀察欄向量間的二元關係:
因此
三欄 線性相關,所以其行列式為 ,是偶數。
同理,
故 產生偶數 minor。
此外,
故 也產生偶數 minor。
最後,
所以 亦線性相關。其整數行列式為
同樣是偶數。
選項分析
第 15 題3 分
- (3%) The board game Set uses a deck of 27 cards.
Each card has three attributes:
• Its shape can be ., *, or #.
• Its size can be small, medium, or large.
• Its number can be one, two, or three.
A triple in this game is three cards such that
• either all shapes are the same or all different,
• either all sizes are the same or all different, and
• either all numbers are the same or all different.
For instance,
🖼️【此處有附圖,請對照原卷】
is a triple because all shapes are different, all sizes are the same (small), and all numbers are different. A subset of cards is said to be triple-fruitful if for any two cards in the subset, the third card that forms a triple with them is also in the subset. What are the possible sizes of a triple-fruitful subset of cards?
(a) 1
(b) 3
(c) 5
(d) 7
(e) 9
(f) 11
(g) 13
(h) 15 (i) 17 (j) 19 (k) 21 (l) 23 (m) 25 (n) 27 (o) 29
登入後即可作答並保存紀錄。
在將卡片屬性編碼後,可將此遊戲的 27 張卡片對應至體 上的三維仿射空間 中的點。
在此空間中,三張卡片 組成一個 triple,若且唯若它們在仿射空間中三點共線,即滿足:
根據定義,子集 稱為 triple-fruitful,代表若 (且 ),則唯一與其共線的第三點 亦屬於 。這表示 在幾何上「包含通過其中任意兩點的整條直線」。
第 16 題4 分
- (4%) Let be a symmetric matrix with real entries. Reorder the following choices to build a proof of the property that has real eigenvalues.
(a) Since is symmetric, .
(b) Since is lower-triangular as well as upper-triangular, it must be diagonal.
(c) Let be the Schur decomposition of , where is lower triangular and is unitary.
(d) Therefore, the diagonals of are the eigenvalues of , and they are equal to their complex conjugates.
登入後即可作答並保存紀錄。
核心觀念
本題考查三個觀念:
-
Schur 分解:任意複數方陣皆可寫成
其中 為 unitary matrix, 為三角矩陣。本題採用下三角形式。 -
實對稱矩陣也是 Hermitian 矩陣:因為 為實矩陣且對稱,
-
三角矩陣的對角線元素就是特徵值:三角矩陣的特徵多項式為
因此其對角線元素即為全部特徵值。
解題方法
先對 作 Schur 分解:
其中 為下三角矩陣、 為 unitary matrix。兩側左乘 、右乘 ,得到
由於 為實對稱矩陣,所以 。因此
另一方面,
故
因為 原本是下三角矩陣,而 是其共軛轉置,必為上三角矩陣。又由 可知 同時是下三角與上三角矩陣,因此 必為對角矩陣。
最後, 表示 與 unitary 相似,兩者具有相同特徵值;而 的對角線元素就是其特徵值。又因為
所以每個對角元素滿足
故 為實數。因此 的所有特徵值皆為實數。
第 17 題4 分
- (4%) Recall that, in Gaussian elimination, we find a row with a nonzero entry on the first column, move this row to the top, and use this row to eliminate all the other entries in the first column. We then ignore the first row and the first column and repeat the same process. In numerical linear algebra, we often do pivoting to reduce rounding errors. That is, when choosing the next row to move to the top, we choose the row with the largest (after taking absolute value) entry on the first column. Let's apply pivoting to the following matrix:
🖼️【此處有附圖,請對照原卷】
(a) -3 9 -2 4
(b) 3 -3 6 2
(c) 6 2 4 2
(d) -3 1 6 4
Reorder the choices above to indicate their order of being selected as the row that maximizes the absolute value of the pivot.
登入後即可作答並保存紀錄。
核心觀念
本題考查數值線性代數(Numerical Linear Algebra)中的部分樞紐高斯消去法(Gaussian Elimination with Partial Pivoting)。
- 高斯消去法(Gaussian Elimination):透過列運算(Row Operations)將矩陣化簡為上三角矩陣(Upper Triangular Matrix)。
- 部分樞紐(Partial Pivoting):在每一步消去前,於目前考慮的樞紐行(Pivot Column)中,尋找從目前列至底部的所有元素中「絕對值最大者」,將該列交換(Row Swap)至目前的主列位置作為樞紐列,藉此降低浮點數運算時的捨入誤差(Rounding Error),避免除以極小數值產生數值不穩定。
解題方法
給定題目中的 4 條列向量,構成 矩陣:
題目給予四個列的代號分別為:
- (a)
- (b)
- (c)
- (d)
要求依據 Partial Pivoting 依序選取樞紐列的先後順序進行排序。
步驟 1:選擇第 1 個樞紐列(第 1 行消去)
觀察第 1 行(Column 1)的所有元素:
- 列 (a):
- 列 (b):
- 列 (c):
- 列 (d):
第一行中絕對值最大者為列 (c) 的 。因此第 1 個被選為樞紐列的是 (c)。
將列 (c) 交換至第 1 列,並使用列 (c) 作為樞紐對其餘列進行消去:
- 對列 (a):
- 對列 (b):
- 對列 (d):
步驟 2:選擇第 2 個樞紐列(第 2 行消去)
排除已固定的列 (c),觀察剩餘列 (a)、(b)、(d) 在第 2 行(Column 2)的數值:
- 列 (a):
- 列 (b):
- 列 (d):
第 18 題4 分
- (4%) Strassen is a famous algorithm for matrix multiplication. Let and be two matrices. And let be , that is,
Note that the preceding equations require eight multiplications. Strassen, instead, recommends computing the following seven products:
(a) Express as a linear combination of .
(b) Express as a linear combination of .
(c) Express as a linear combination of .
(d) Express as a linear combination of .
登入後即可作答並保存紀錄。
(a)
(b)
(c)
(d)
第 19 題5 分
- Denote the bus waiting time by a continuous random variable . On a rainy day, the bus waiting time is uniformly distributed on the interval . On a non-rainy day, the bus waiting time is uniformly distributed on the interval . The probability that a given day is rainy is .
(a) (5%) Derive the PDF . (Please express your answer in fractional form for full credit.)
(b) (5%) Compute .
(c) (5%) Compute . (Please express your answer in fractional form for full credit.)
登入後即可作答並保存紀錄。
核心觀念
本題考查「混合型連續隨機變數」:
-
雨天與非雨天分別具有不同的條件機率密度函數。
-
利用全機率法則求 的邊際 PDF:
-
期望值使用全期望公式:
-
變異數使用全變異數公式:
其中 表示下雨。
解題方法
雨天機率為
非雨天機率為
(a) 求 PDF
雨天時,
因此
非雨天時,
因此
依照 的範圍分段:
當
兩種天氣皆可能,因此
當
雨天不可能等待超過 ,只有非雨天有密度:
其他範圍
所以
端點的函數值不影響連續型隨機變數的機率。
(b) 求
均勻分布 的期望值為
雨天的條件期望:
非雨天的條件期望:
使用全期望公式:
第 20 題5 分
- (5%) is a Poisson random variable with mean 5. Please compute .
登入後即可作答並保存紀錄。
核心觀念
若 ,其機率質量函數為
本題要求計算 ,可使用 Poisson 隨機變數的機率生成函數:
題目給定平均數為 ,因此 。
解題方法
直接代入 Poisson 分配的機率生成函數:
也可由期望值定義驗證:
第 21 題5 分
- The joint PMF of M and J is given as
for and . Otherwise, . Let denote the event that .
(a) (5%) Derive the marginal PMF .
(b) (5%) Derive the conditional PMF .
(c) (5%) Derive .
登入後即可作答並保存紀錄。
(a) 邊際 PMF
其中 為第 個調和數。
(b) 條件 PMF 於事件
因此
(c) 條件期望
第 22 題5 分
- (5%) Mike and Tom have an appointment this afternoon. Mike arrives uniformly at random between 2:00 PM and 3:00 PM, and Tom arrives uniformly at random between 2:30 PM and 3:30 PM. Assume their arrival times are independent. The person who arrives first will wait for at most 15 minutes for the other person before leaving. Find the probability that Mike and Tom meet (i.e., that they are both present at the same time). You can express your answer in fractional form.
登入後即可作答並保存紀錄。
解題步驟
-
設 為 Mike 到達的分鐘數(以 2:00 為 0),
為 Tom 到達的分鐘數(以 2:00 為 0)。
兩變數獨立,樣本空間為長方形
面積 。 -
先到者最多等 分鐘,等同於兩人相差不超過 分鐘:
-
對每個 ,允許的 範圍為
於 分段計算可得
第 23 題5 分
- Let be i.i.d. exponential random variables with mean 2. Let .
(a) (5%) Derive the MGF , and state the valid range of .
(b) (5%) Compute .
登入後即可作答並保存紀錄。
核心觀念
本題考查:
- 指數分布的參數轉換與 MGF。
- 獨立隨機變數和的 MGF:
- 線性組合的變異數:
其中各隨機變數彼此獨立。
已知 為平均數 的指數隨機變數,因此其速率參數為
所以
解題方法
(a) 求 的 MGF 與有效範圍
指數分布 的 MGF 為
代入 ,得到
由
且 相互獨立,因此
將指數拆開:
分別計算: