Data Structures and Algorithms CMPSC 465

Transcription

1 Data Structures and Algorithms CMPSC 465 LECTURES 7-8 More Divide and Conquer Multiplication Adam Smith S. Raskhodnikova and A. Smith; based on slides by E. Demaine and C. Leiserson

2 John McCarthy ( ) Pioneer in artificial intelligence In fact, he coined the term Lots of work on how reasoning can be automated Meta-lesson: you learn a lot about something when you teach a computer to do it.

3 McCarthy s 91 function Why be so careful about correctness with recursive procedures? M(n) = { n 10, if n>100 M(M(n + 11)), if n 100 This function is tricky to reason about because it doesn t always make recursive calls on smaller inputs Programmer s goal: code that is easy to understand Do not write circular programs! S. Raskhodnikova and A. Smith; based on slides by E. Demaine and C. Leiserson 3

4 Reminder: geometric series x x x = for x < x x x x x n n = for x 1

5 Review For each tree: How many leaves? How many nodes in total? (big-theta) H

6 Multiplying large integers Given n-bit integers a, b (in binary), compute c=ab Naive (grade-school) algorithm: a n-1 a n-2 a 0 Write a,b in binary Compute n intermediate products Do n additions Running time? Total work: (n 2 ) n bits b n-1 b n-2 b 0 n bits n bits 2n bit output

7 Divide-and-conquer design 1. Divide the problem (instance) into subproblems. 2. Conquer the subproblems by solving them recursively. 3. Combine subproblem solutions.

8 Multiplying large integers Divide and Conquer (Attempt #1): Write a = A 1 2 n /2 + A 0 b = B 1 2 n /2 + B 0 We want ab = A 1 B 1 2 n + (A 1 B 0 + B 1 A 0 ) 2 n /2 + A 0 B 0 Multiply n/2 bit integers recursively T(n) = 4T(n/2) + (n) Alas! this is still (n 2 ). (Exercise: write out the recursion tree.)

9 Multiplying large integers Divide and Conquer (Attempt #1): Write a = A 1 2 n /2 + A 0 b = B 1 2 n /2 + B 0 We want ab = A 1 B 1 2 n + (A 1 B 0 + B 1 A 0 ) 2 n /2 + A 0 B 0 Multiply n/2 bit integers recursively T(n) Consider = 4T(n/2) the expression + (n) (A Alas! this is still (n 2 0 +A 1 ) (B 0 + B 1 ) = A 0 ) B. 0 + A 1 B 1 + (A 0 B 1 + B 1 A 0 ) (Exercise: We can get write away out with the 3 recursion multiplications! tree.) (in yellow) x = A 1 B 1 y = A 0 B 0 z = (A 0 +A 1 )(B 0 +B 1 ) Now we use ab = A 1 B 1 2 n + (A 1 B 0 + B 1 A 0 ) 2 n /2 + A 0 B 0 = x 2n + (z x y) 2 n /2 + y

10 Multiplying large integers MULTIPLY (n, a, b) a and b are n-bit integers Assume n is a power of 2 for simplicity 1. If n 2 then use grade-school algorithm else 2. A 1 a div 2 n /2 ; B 1 b div 2 n /2 ; 3. A 0 a mod 2 n /2 ; B 0 b mod 2 n /2. 4. x MULTIPLY(n/2, A 1, B 1 ) 5. y MULTIPLY(n/2, A 0, B 0 ) 6. z MULTIPLY(n/2, A 1 +A 0, B 1 +B 0 ) 7. Output x 2 n + (z x y)2 n /2 + y

11 Multiplying large integers The resulting recurrence T(n) = 3T(n/2) + (n) We will see that T(n) = (n log 23 ) = (n 1.59 ) Note: There is a (n log n) algorithm for multiplication (assumes constant-time word ops) Actually, (n log(n) 2 log*(n) ) bit-level operations (due to Martin Fürer, 2007)

12 Reminder: recursion trees Technique for guessing solution to recurrences Write out tree of recursive calls Each node gets assigned the work done during that call to the procedure (dividing and combining) Total work is sum of work at all nodes

13 Recursion tree for multiplication Solve T(n) = 3T(n/2) + cn, where c > 0 is constant. cn cn lg 2 n cn/2 cn/4 cn/4 (1) cn/4 cn/2 cn/2 At bottom level: (3/2)cn (9/4)cn At level k: (3/2) k cn (3/2) log n cn = c n log 2 3

14 Recursion tree for multiplication lg 2 n Solve T(n) = 3T(n/2) + cn, where c > 0 is constant. (1) cn/2 cn/4 cn/4 cn/4 Total work log 2 n = k= 0 cn cn/2 3 2 k cn cn/2 = (3 log 2 n ) = ( n log cn (3/2)cn (9/4)cn Geometric series: When base of exponent is constant (e.g. 3/2), the largest term is a constant fraction of total 2 3 )

15 Divide-and-conquer design 1. Divide the problem (instance) into subproblems. 2. Conquer the subproblems by solving them recursively. 3. Combine subproblem solutions.

16 Binary search Find an element in a sorted array: 1. Divide: Check middle element. 2. Conquer: Recursively search 1 subarray. 3. Combine: Trivial. Example: Find

17 Binary search Find an element in a sorted array: 1. Divide: Check middle element. 2. Conquer: Recursively search 1 subarray. 3. Combine: Trivial. Example: Find

18 Binary search Find an element in a sorted array: 1. Divide: Check middle element. 2. Conquer: Recursively search 1 subarray. 3. Combine: Trivial. Example: Find

19 Binary search Find an element in a sorted array: 1. Divide: Check middle element. 2. Conquer: Recursively search 1 subarray. 3. Combine: Trivial. Example: Find

20 Binary search Find an element in a sorted array: 1. Divide: Check middle element. 2. Conquer: Recursively search 1 subarray. 3. Combine: Trivial. Example: Find

21 Binary search Find an element in a sorted array: 1. Divide: Check middle element. 2. Conquer: Recursively search 1 subarray. 3. Combine: Trivial. Example: Find

22 Binary Search BINARYSEARCH(b, A[1.. n]) find b in sorted (increasing order) array A 1. If n=0 then return not found 2. If A[ n/2 ] = b then return n/2 3. If A[ n/2 ] > b then 4. return BINARYSEARCH(A[1.. n/2 ]) 5. Else 6. return n/2 + BINARYSEARCH(A[ n/ n])

23 Proof of correctness Omitted here, similar to Merge-Sort

24 Recurrence for binary search T(n) = 1 T(n/2) + (1) # subproblems subproblem size work dividing and combining

25 Recurrence for binary search T(n) = 1 T(n/2) + (1) # subproblems subproblem size work dividing and combining T(n) = T(n/2) + c = T(n/4) + 2c = c log n = (lg n).

26 Exponentiation Problem: Compute a b, where b N is n bits long Question: How many multiplications? Naive algorithm: (b) = (2 n ) (exponential in the input length!) Divide-and-conquer algorithm: a b = a b/2 a b/2 if b is even; a (b 1)/2 a (b 1)/2 a if b is odd. T(b) = T(b/2) + (1) T(b) = (log b) = (n).

27 So far: 3 recurrences Merge Sort T(n) = 2 T(n/2) + (n) = (n log n) Binary Search / Exponentiation T(n) = 1 T(n/2) + (1) = (log n) Multiplication T(n) = 3 T(n/2) + (n) = (n log 2 3 ) Coming soon: systematic ways to solve these.