4.2 Strassen's algorithm for matrix multiplication
4.2-1¶
Use Strassen's algorithm to compute the matrix product
$$ \begin{pmatrix} 1 & 3 \\ 7 & 5 \end{pmatrix} \begin{pmatrix} 6 & 8 \\ 4 & 2 \end{pmatrix} . $$
Show your work.
The first matrices are
$$ \begin{array}{ll} S_1 = 6 & S_6 = 8 \\ S_2 = 4 & S_7 = -2 \\ S_3 = 12 & S_8 = 6 \\ S_4 = -2 & S_9 = -6 \\ S_5 = 6 & S_{10} = 14. \end{array} $$
The products are
$$ \begin{aligned} P_1 & = 1 \cdot 6 = 6 \\ P_2 & = 4 \cdot 2 = 8 \\ P_3 & = 6 \cdot 12 = 72 \\ P_4 & = -2 \cdot 5 = -10 \\ P_5 & = 6 \cdot 8 = 48 \\ P_6 & = -2 \cdot 6 = -12 \\ P_7 & = -6 \cdot 14 = -84. \end{aligned} $$
The four matrices are
$$ \begin{aligned} C_{11} & = 48 + (-10) - 8 + (-12) = 18 \\ C_{12} & = 6 + 8 = 14 \\ C_{21} & = 72 + (-10) = 62 \\ C_{22} & = 48 + 6 - 72 - (-84) = 66. \end{aligned} $$
The result is
$$ \begin{pmatrix} 18 & 14 \\ 62 & 66 \end{pmatrix} . $$
4.2-2¶
Write pseudocode for Strassen's algorithm.
STRASSEN(A, B)
n = A.rows
if n == 1
return a[1, 1] * b[1, 1]
let C be a new n × n matrix
A[1, 1] = A[1..n / 2][1..n / 2]
A[1, 2] = A[1..n / 2][n / 2 + 1..n]
A[2, 1] = A[n / 2 + 1..n][1..n / 2]
A[2, 2] = A[n / 2 + 1..n][n / 2 + 1..n]
B[1, 1] = B[1..n / 2][1..n / 2]
B[1, 2] = B[1..n / 2][n / 2 + 1..n]
B[2, 1] = B[n / 2 + 1..n][1..n / 2]
B[2, 2] = B[n / 2 + 1..n][n / 2 + 1..n]
S[1] = B[1, 2] - B[2, 2]
S[2] = A[1, 1] + A[1, 2]
S[3] = A[2, 1] + A[2, 2]
S[4] = B[2, 1] - B[1, 1]
S[5] = A[1, 1] + A[2, 2]
S[6] = B[1, 1] + B[2, 2]
S[7] = A[1, 2] - A[2, 2]
S[8] = B[2, 1] + B[2, 2]
S[9] = A[1, 1] - A[2, 1]
S[10] = B[1, 1] + B[1, 2]
P[1] = STRASSEN(A[1, 1], S[1])
P[2] = STRASSEN(S[2], B[2, 2])
P[3] = STRASSEN(S[3], B[1, 1])
P[4] = STRASSEN(A[2, 2], S[4])
P[5] = STRASSEN(S[5], S[6])
P[6] = STRASSEN(S[7], S[8])
P[7] = STRASSEN(S[9], S[10])
C[1..n / 2][1..n / 2] = P[5] + P[4] - P[2] + P[6]
C[1..n / 2][n / 2 + 1..n] = P[1] + P[2]
C[n / 2 + 1..n][1..n / 2] = P[3] + P[4]
C[n / 2 + 1..n][n / 2 + 1..n] = P[5] + P[1] - P[3] - P[7]
return C
4.2-3¶
How would you modify Strassen's algorithm to multiply $n \times n$ matrices in which $n$ is not an exact power of $2$? Show that the resulting algorithm runs in time $\Theta(n^{\lg7})$.
We can just extend it to an $n \times n$ matrix and pad it with zeroes. It's obviously $\Theta(n^{\lg7})$.
4.2-4¶
What is the largest $k$ such that if you can multiply $3 \times 3$ matrices using $k$ multiplications (not assuming commutativity of multiplication), then you can multiply $n \times n$ matrices is time $o(n^{\lg 7})$? What would the running time of this algorithm be?
Assume $n = 3^m$ for some $m$. Then, using block matrix multiplication, we obtain the recursive running time $T(n) = kT(n / 3) + O(1)$.
By master theorem, we can find the largest $k$ to satisfy $\log_3 k < \lg 7$ is $k = 21$.
4.2-5¶
V. Pan has discovered a way of multiplying $68 \times 68$ matrices using $132464$ multiplications, a way of multiplying $70 \times 70$ matrices using $143640$ multiplications, and a way of multiplying $72 \times 72$ matrices using $155424$ multiplications. Which method yields the best asymptotic running time when used in a divide-and-conquer matrix-multiplication algorithm? How does it compare to Strassen's algorithm?
Using what we know from the last exercise, we need to pick the smallest of the following
$$ \begin{aligned} \log_{68} 132464 & \approx 2.795128 \\ \log_{70} 143640 & \approx 2.795122 \\ \log_{72} 155424 & \approx 2.795147. \end{aligned} $$
The fastest one asymptotically is $70 \times 70$ using $143640$.
4.2-6¶
How quickly can you multiply a $kn \times n$ matrix by an $n \times kn$ matrix, using Strassen's algorithm as a subroutine? Answer the same question with the order of the input matrices reversed.
- $(kn \times n)(n \times kn)$ produces a $kn \times kn$ matrix. This produces $k^2$ multiplications of $n \times n$ matrices.
- $(n \times kn)(kn \times n)$ produces an $n \times n$ matrix. This produces $k$ multiplications and $k - 1$ additions.
4.2-7¶
Show how to multiply the complex numbers $a + bi$ and $c + di$ using only three multiplications of real numbers. The algorithm should take $a$, $b$, $c$ and $d$ as input and produce the real component $ac - bd$ and the imaginary component $ad + bc$ separately.
The three matrices are
$$ \begin{aligned} A & = (a + b)(c + d) = ac + ad + bc + bd \\ B & = ac \\ C & = bd. \end{aligned} $$
The result is
$$(B - C) + (A - B - C)i.$$