Input Decidable Language -- Program Halts on all Input Encoding of Input -- Natural Numbers Encoded in Binary or Decimal, Not Unary

Transcription

1 Complexity Analysis

2 Complexity Theory Input Decidable Language -- Program Halts on all Input Encoding of Input -- Natural Numbers Encoded in Binary or Decimal, Not Unary Output TRUE or FALSE Time and Space Problems How much Time does the Algorithm Take to Execute? How much Memory is Required to Execute the Algorithm? As Functions of the Size of the Input

3 Examples Hamiltonian Circuit -- Decision Problem Input = Matrix Describing a Graph Output = YES if Graph has a Hamiltonian Circuit = NO if Graph does not have a Hamiltonian Circuit Traveling Salesman -- Optimization Problem Input = Matrix Describing the Cost of Each Edge Max Cost = Maximal Allowable Cost of Circuit Output = YES if Graph has Circuit with Cost Max Cost = NO if Graph does not have a Circuit with Cost Max Cost

4 Complexity and Turing Machines One Tape Deterministic Turing Machines Time Complexity = Number of Commands Executed Space Complexity = Number of Tape Squares Visited Multiple Tape Turing Machines and Actual Digital Computers Complexity Differs from Single Tape Turing Machines by Only a Polynomial Factor Non-Deterministic Turing Machines Require Different Complexity Analysis from Deterministic Turing Machines

5 Complexity of Turing Machines Setup M = Turing Machine that Halts on All Input n = Size of Input Time and Space Complexity Worst Case Analysis t M : N N -- t M (n) = Maximal Time M Takes to Execute on a Problem of Size n s M : N N -- s M (n) = Maximal Memory Required by M to Execute a Problem of Size n For Non-Deterministic Machines, Consider All Possible Paths

6 Asymptotic Complexity Classes Big O -- Upper Bound Big Ω -- Lower Bound Small o -- Much Bigger Bound Small ω -- Much Smaller Bound Big Theta Θ -- Tight Bound

7 Big O and Big Ω Big O -- Upper Bound f = O(g) f O(g) f is bounded above by g O(g) = {h h(n) C h g(n) for all n k h for some constant C h } Big Ω -- Lower Bound f = Ω(g) f Ω(g) f is bounded below by g Ω(g) = {h h(n) C h g(n) for all n k h for some constant C h } Observations k and C are not unique g is not unique

8 Big O and Big Ω and Limits Limits lim n f (n) g(n) < f O(g) lim n f (n) g(n) > 0 f Ω(g)

9 Small o and Small ω Definitions f is o(g) lim n f (n) g(n) = 0 (much smaller) f is ω (g) lim n f (n) g(n) = (much bigger) Observations f is o(g) f is O(g) (much smaller smaller) f is ω (g) f is Ω(g) (much bigger bigger)

10 Big-Θ -- Tight Bound Definition f = Θ(g) f Θ(g) f is bounded above and below by a constant times g Θ(g) = {h Dg(h) h(n) C h g(n) for all n k h for some constants C h, D h } Meaning f Θ(g) f Ω(g) and f O(g) Observations k, c, C are not unique g is not unique Limits 0 < lim n f (n) g(n) < f Θ(g)

11 Most Important Big O is the most important complexity class, since we are most interested in an upper bound on the time of an algorithm.

12 Big-O Definition f is O(g) means there are numbers k and C such that -- f (n) Cg(n) for all n > k. Meaning For all sufficiently large integers, f (n) is less than a constant multiple of g(n). g is only an Upper Bound on the size of f. Observations k and C are not unique. g is not unique. g is usually chosen to be well-known function: log(n), n p, 2 n.

13 Examples 1. c p n p + + c 1 n + c 0 = O(n p ) 2. c p n p + + c 1 n + c 0 d q n q + + d 1 n + d 0 = O(n p q ) n 3. k p = O(n p+1 ) k =1 n 4. 2 p = O(2 n ) p=1 5. n! = O(n n )

14 Most Common Orders 1, log(n), n, nlog(n), n 2, n p, 2 p, n!, n n Constant, Logarithmic, Linear, Polynomial, Exponential

15 Properties of Big O 0. f = O( f ) 1. f = O(g) and g = O(h) f = O(h) 2. f = O(g) f + c = O(g) 3. f = O(g) c f = O(g) 4. f 1 = O(g 1 ) and f 2 = O(g 2 ) f 1 + f 2 = O ( max(g 1, g 2 )) 5. f 1 = O(g 1 ) and f 2 = O(g 2 ) f 1 f 2 = O(g 1 g 2 )

16 Properties of Big O (continued) FALSE Property 2. f = O(g) f + c = O(g) Example 1 n 2 = O 1 n 1 n 2 +1 O 1 n True Property 2*. f = O(g) and 1 = O( f ) f + c = O(g)

17 Properties of Big O for Special Functions 6. p q n p = O(n q ) 7. c p n p + + c 1 n + c 0 = O(n p ) 8, a < b a n = O(b n ) 9. a,b > 1 O ( Log a (n)) = O ( Log b (n)) -- Log a (n) = Log a (b)log b (n) 10. O(n p ) O ( n p Log b (n)) O(n p +1 ) n = O(n!) 12. n! = O(n n )

18 Examples Find Simple O(g) estimates for the following functions: 1. (n 2 + 8)(n +1) 2. (nlog(n) + n 2 )(n 3 + 3) 3. ( n!+ 2 n )( n 3 + log(n 2 +1) )

19 True or False Prove or give a counterexample For all positive functions f and g, either f = O(g) or g = O( f ).

20 True or False Prove or give a counterexample For all positive functions f and g, either f = O(g) or g = O( f ). Counterexample f (n) = 1+ ncos 2 nπ 2 g(n) = 1 + nsin 2 n π 2

21 Examples of o( f ) lim n ln(n) n = 0 ln(n) is o(n) ( ln(n) ) p lim n n = 0 ( ln(n) ) p is o(n) lim n n p ln(n) n p +1 = 0 n p ln(n) is o(n p+1 ) lim n n p 2 n = 0 n p is o(2 n ) lim n P(n) 2 n = 0 P(n) is o(2 n ), P(n) Polynomial

22 Examples of Θ( f ) 1. c nn p + + c 1 n + c 0 = Θ(n p ) 2. c p n p + + c 1 n + c 0 d q n q + + d 1 n + d 0 = Θ(n p q ) n 3. k p = Θ(n p +1 ) (Hint: Integrate) k =1 n 4. 2 p = Θ(2 n ) p=1

23 Recursion vs. Dynamic Programming Polynomial Interpolation Neville s Algorithm (Time and Space) O(n 2 ) vs. O(2 n ) Fibonacci Numbers Memoization O(n) vs. O(2 n )

24 Neville s Algorithm (Polynomial Interpolation) -- Recursive Implementation I 1,,n (x) I 1,,n 1 (x) I 2,,n (x) I 1,2 (x) I 1,,n 2 (x) I 2,,n 1 (x) I 2,,n 1(x) I 3,,n 1(x) I n 1, n (x) y 1 y 2 y n 1 y n n 1 Levels 2 n 2 = O(2 n ) multiplications

25 Neville s Algorithm (Polynomial Interpolation) -- Iterative Implementation I 1, 2, 3, 4 I 1, 2, 3 I 2, 3, 4 I 1,2 I 2, 3 I 3, 4 y 1 y 2 y 3 y 4 n 1 n 1 Levels 2 k = (n 1)n = O(n 2 ) multiplications k =1

26 Fibonacci Recurrence f n f n 1 f n 2 f 3 f n 2 f n 3 f n 3 f n 4 f 3 f 1 f 2 f 1 f 2 f n = f n 1 + f n 2 T ( f n ) = T ( f n 1 ) +T ( f n 2 )

27 Fibonacci Numbers -- Memoization f 1 f 3 f 5 f 2n 3 f 2n 1 f 2 f 4 f 6 f 2n 2 f 2n T ( f 2n ) = 2n 2 = O(n) Additions

28 Polynomial Time Algorithms Polynomial Evaluation Horner s Method (Time) O(n) vs. O(n 2 ) Polynomial Multiplication Fast Fourier Transform (FFT) O ( n log(n) ) vs. O(n 2 ) Matrix Multiplication Strassen s Algorithm O n 2.8 ( ) vs. O(n 3 )

29 Horner s Method Example 7x 3 3x 2 +11x multiplies ((7x 3)x +11)x multiplies General Case a nx n + a n 1 x n 1 n + + a 1 x + a 0 -- k = k=0 n(n +1) 2 = O(n 2 ) multiplies ((a n x + a n 1 )x + + a 1 )x + a 0 -- n = O(n) multiplies

30 Searching Algorithms Linear Search Worst Case Analysis -- O(n) Average Case Analysis -- O(n) Binary Search -- O ( Log(n) ) Root Finding (Mathematica Code) Ray Tracing

31

32

33

34

35 Example Integer Division Input: m, n = integers Output: q, r = integers (quotient and remainder) -- n = mq + r with 0 r < m Algorithm q = 0 r = n While r m q = q +1 Loop Invariant n = mq + r r = r m

36 Proofs Proof of Loop Invariance By induction on the number of iteration of the loop. Base Case: q = 0 and r = n n = mq + r. Induction: Suppose that n = mq + r after k iterations of the loop. Proof of Program Correctness Must show that n = mq + r after k +1 iterations of the loop. But after k +1 iterations of the loop: (mq + r) k +1 = mq k +1 + r k+1 = m (q k +1) + r k m = m q k + r k (inductive hypothesis) = n. 1. The loop will terminate with r < m. 2. When the loop terminates n = m q + r. (Loop Invariance)

37 Complexity of Integer Division Worst Case: m = 2, n odd O(n / 2)

38 Example GCD -- Greatest Common Divisor Input: m, n = integers Output: GCD(m,n) Euclidean Algorithm x = m y = n While y 0 Loop Invariant r = remainder of x divided by y GCD(x, y) = GCD(m, n) (use an efficient division algorithm) x = y y = r Output: x is GCD(m, n)

39 Proof of Loop Invariance By induction on the number of iteration of the loop. Base Case: x = m and y = n GCD(x, y) = GCD(m, n). Induction: Suppose that GCD(x, y) = GCD(m, n) after k iterations of the loop. Must show that GCD(x, y) = GCD(m, n) after k +1 iterations. After the first line of the (k +1) st iteration of the loop: x = yq + r r = x yq GCD(y, r) = GCD(x, y) and by the inductive hypothesis GCD(x, y) = GCD(m, n) so GCD(y, r) = GCD(m, n). But after the (k +1) st iteration of the loop: x = y and y = r so GCD(x, y) = GCD(y, r) = GCD(m, n).

40 Complexity GCD Algorithm Standard GCD Algorithm Factor m and n into Prime Factors Find Common Prime Factors No Efficient Algorithm for Step 1 Euclid s GCD Algorithm m n / 2: (n, m) (m,r) r < n / 2 m < n / 2: (n, m) (m,r) m < n / 2 -- Second Parameter Decreases by at Least 1/ 2 in at Most 2 Step -- O ( Log 2 (n))

41 Searching Algorithms Depth First Less Space -- 1 Path at a Time More Time -- May Explore Bad Path Breadth First More Space -- Stores all Paths Simultaneously Less Time -- May Find Good Path and Halt Iterative Deepening All Paths to Depth 1, All Paths to Depth 2, Space Complexity = Space Complexity of Depth First Search Time Complexity Breadth First Search Compromise Between Depth First and Breadth First Search