Chain Matrix Multiplication

Dynamic Programming Formulation

• We will store the solutions to the subproblem
in a table and build the table bottom-up
(why)?
• For 1 ≤ i ≤ j ≤ n, let m[i, j] denote the
minimum number of multiplications needed to
compute Ai..j .
• The optimum can be described by the
following recursive formulation.

Algorithms – p. 369
Chain Matrix Multiplication
• If i = j, there is only one matrix and thus
m[i, i] = 0 (the diagonal entries).
• If i < j, the we are asking for the product Ai..j .
• This can be split by considering each k,
i ≤ k < j, as Ai..k times Ak+1..j .

Algorithms – p. 370
Chain Matrix Multiplication
• The optimum time to compute Ai..k is m[i, k]
and optimum time for Ak+1..j is in m[k + 1, j].
• Since Ai..k is a pi−1 × pk matrix and Ak+1..j is
pk × pj matrix, the time to multiply them is
pi−1 × pk × pj .

Algorithms – p. 371
Chain Matrix Multiplication
• This suggests the following recursive rule:
m[i, i] = 0
m[i, j] = min (m[i, k] + m[k + 1, j] + pi−1 pk pj )
i≤k<j

Algorithms – p. 372
Chain Matrix Multiplication
• We do not want to calculate m entries
recursively.
• How should we proceed?
• We will fill m along the diagonals. Here is
how.

Algorithms – p. 373
Chain Matrix Multiplication
• Set all m[i, i] = 0 using the base condition.
• Compute cost for multiplication of a sequence
of 2 matrices.
• These are
m[1, 2], m[2, 3], m[3, 4], . . . , m[n − 1, n].
• m[1, 2], for example is
m[1, 2] = m[1, 1] + m[2, 2] + p0 · p1 · p2

Algorithms – p. 374
Chain Matrix Multiplication

For example, for m for product of 5 matrices at

this stage would be:
←m[1, 2]
m[1, 1]
↓
←m[2, 3]
m[2, 2]
↓
←m[3, 4]
m[3, 3]
↓
←m[4, 5]
m[4, 4]
↓
m[5, 5]

Algorithms – p. 375
Chain Matrix Multiplication
• Next, we compute cost of multiplication for
sequences of three matrices.
• These are
m[1, 3], m[2, 4], m[3, 5], . . . , m[n − 2, n].
• m[1, 3], for example is

m[1, 1] + m[2, 3] + p0 · p1 · p3
m[1, 3] = min
m[1, 2] + m[3, 3] + p0 · p2 · p3

Algorithms – p. 376
Chain Matrix Multiplication
• We repeat the process for sequences of four,
five and higher number of matrices.
• The final result will end up in m[1, n].
• Let us go through an example. We want to
find the optimal multiplication order for
A1 · A2 · A3 · A4 · A5
(5×4) (4×6) (6×2) (2×7) (7×3)

Algorithms – p. 377
Chain Matrix Multiplication

Initially

0
0
0
0
0

Algorithms – p. 378
Chain Matrix Multiplication

m[1, 2] = m[1, 1] + m[2, 2] + p0 · p1 · p2

= 0 + 0 + 5 · 4 · 6 = 120
m[2, 3] = m[2, 2] + m[3, 3] + p1 · p2 · p3
= 0 + 0 + 4 · 6 · 2 = 48

Algorithms – p. 379
Chain Matrix Multiplication

m[3, 4] = m[3, 3] + m[4, 4] + p2 · p3 · p4

= 0 + 0 + 6 · 2 · 7 = 84
m[4, 5] = m[4, 4] + m[5, 5] + p3 · p4 · p5
= 0 + 0 + 2 · 7 · 3 = 42

Algorithms – p. 380
Chain Matrix Multiplication

after product of 2 matrices (first super diagonal)

0 120
0 48
0 84
0 42
0

Algorithms – p. 381
Chain Matrix Multiplication

m[1, 3] = m[1, 1] + m[2, 3] + p0 · p1 · p3 =

= 0 + 48 + 5 · 4 · 2 = 88
m[1, 3] = m[1, 2] + m[3, 3] + p0 · p2 · p3
= 120 + 0 + 5 · 6 · 2 = 180
minimum m[1, 3] = 88 at k = 1

Algorithms – p. 382
Chain Matrix Multiplication

m[2, 4] = m[2, 2] + m[3, 4] + p1 · p2 · p4

= 0 + 84 + 4 · 6 · 7 = 252
m[2, 4] = m[2, 3] + m[4, 4] + p1 · p3 · p4
= 48 + 0 + 4 · 2 · 7 = 104
minimum m[2, 4] = 104 at k = 3

Algorithms – p. 383
Chain Matrix Multiplication

m[3, 5] = m[3, 3] + m[4, 5] + p2 · p3 · p5

= 0 + 42 + 6 · 2 · 3 = 78
m[3, 5] = m[3, 4] + m[5, 5] + p2 · p4 · p5
= 84 + 0 + 6 · 7 · 3 = 210
minimum m[3, 5] = 78 at k = 3

Algorithms – p. 384
Chain Matrix Multiplication

product of 3 matrices, (second super diagonal)

0 120 88
0 48 104
0 84 78
0 42
0

Algorithms – p. 385
Chain Matrix Multiplication

m[1, 4] = m[1, 1] + m[2, 4] + p0 · p1 · p4

= 0 + 104 + 5 · 4 · 7 = 244
m[1, 4] = m[1, 2] + m[3, 4] + p0 · p2 · p4
= 120 + 84 + 5 · 6 · 7 = 414
m[1, 4] = m[1, 3] + m[4, 4] + p0 · p3 · p4
= 88 + 0 + 5 · 2 · 7 = 158
minimum m[1, 4] = 158 at k = 3

Algorithms – p. 386
Chain Matrix Multiplication

m[2, 5] = m[2, 2] + m[3, 5] + p1 · p2 · p5

= 0 + 78 + 4 · 6 · 3 = 150
m[2, 5] = m[2, 3] + m[4, 5] + p1 · p3 · p5
= 48 + 42 + 4 · 2 · 3 = 114
m[2, 5] = m[2, 4] + m[5, 5] + p1 · p4 · p5
= 104 + 0 + 4 · 7 · 3 = 188
minimum m[2, 5] = 114 at k = 3

Algorithms – p. 387
Chain Matrix Multiplication

Product of 4 matrices
0 120 88 158
0 48 104 114
0 84 78
0 42
0

Algorithms – p. 388
Chain Matrix Multiplication
m[1, 5] = m[1, 1] + m[2, 5] + p0 · p1 · p5
= 0 + 114 + 5 · 4 · 3 = 174
m[1, 5] = m[1, 2] + m[3, 5] + p0 · p2 · p5
= 120 + 78 + 5 · 6 · 3 = 288
m[1, 5] = m[1, 3] + m[4, 5] + p0 · p3 · p5
= 88 + 42 + 5 · 2 · 3 = 160
m[1, 5] = m[1, 4] + m[5, 5] + p0 · p4 · p5
= 158 + 0 + 5 · 7 · 3 = 263
minimum m[1, 5] = 160 at k = 3
Algorithms – p. 389
Chain Matrix Multiplication

Final cost matrix

0 120 88 158 160
0 0 48 104 114
0 0 0 84 78
0 0 0 0 42
0 0 0 0 0

Algorithms – p. 390
Chain Matrix Multiplication

Order in which m entries are calculated

0 1 5 8 10
0 0 2 6 9
0 0 0 3 7
0 0 0 0 4
0 0 0 0 0

Algorithms – p. 391
Chain Matrix Multiplication

Split k values

0 1 1 3 3
0 2 3 3
0 3 3
0 4
0

Algorithms – p. 392
Chain Matrix Multiplication

Optimal Order for multiplication

((A1 (A2 A3 ))(A4 A5 ))

1 4

A1 2 A4 A5

A2 A3

Algorithms – p. 393