Algorithm Analysis - Math CheatSheet

Logarithm properties Asymptotic Complexities

• Sometimes we may obtain a simpler expression by
• Definition: base logbase x
=x Assuming: limn→∞ ( fg(n)
(n)
)=L changing variables:
logb (x)
√
• Base change: loga (x) = Example: T (n) = 2 ∗ T ( n) + log(n)
logb (a) • 0 ≤ L < ∞ ⇒ f (n) ∈ O(g(n))
√
• Product rule: logb (x ∗ y) = logb (x) + logb (y) • 0 < L < ∞ ⇒ f (n) ∈ Θ(g(n)) 1. Let n = 2m → m = log(n) → n = 2m/2
2. Let S(m) = T (2m )
• Quotient rule: logb (x/y) = logb (x) − logb (y) • 0 < L ≤ ∞ ⇒ f (n) ∈ Ω(g(n))
• alogB (b) = blogB (a) Then: T (2m ) = 2 ∗ T (2m/2 ) + m
• L = 0 ⇒ f (n) ∈ o(g(n))
→ S(m) = 2 ∗ S(m/2) + m
Common derivatives • L = ∞ ⇒ f (n) ∈ ω(g(n)) → S(m) = θ(m ∗ log(m)) – Master Th.
→ T (n) = T (2m ) = S(m) = θ(m ∗ log(m))
F(x) F(x)’
Master Theorem → T (n) = θ(log(n) ∗ log(log(n)))
xn n ∗ x(n−1) s
bx bx ∗ ln(b) Suppose: T (n) = a ∗ T (n/b) + f (n), T (1) = θ(1). • It may be useful to search for a lower and an upper
1 Subject to: a ≥ 1, b > 1, f is an asymptotically positive
logb (x) x∗ln(b) bound that is easier to compute.
f (x) ∗ g(y) f (x)0 ∗ g(y) + f (x) ∗ g(y)0 function. Plogb (n)−1 i Example: T (n) = T (n/2 − log2 (n)) + 1
f (x) f (x)0 ∗g(y)−f (x)∗g(y)0 Then: T (n) = i=0 a ∗ f (n/bi ) + θ(nlogb (a) )
g(y) g(x)2 Proof using recursion trees: Consider:
0 0
f (g(x)) f (g(x)) ∗ g (x) T1 (n) = T1 (n/4) + 1 → T1 (n) = θ(log(n))
• logb (n) - height of the recursion tree
T2 (n) = T2 (n/2) + 1 → T2 (n) = θ(log(n))
L’Hospital’s Rule
• a ∗ f (n/b ) - cost of all internal nodes at depth i
i i
Since: n/4 ≤ n/2 − log2 (n) ≤ n/2
Assuming: → T1 (n) ≤ T (n) ≤ T2 (n)
• θ(nlogb (a) ) - cost associated with the leaves
• limx→a f (x) and limx→a g(x) are either 0 or ∞ → T (n) ∈ θ(log(n))
0
Master Theorem
• limx→a fg0 (x)
(x)
exists
1. f (n) = O(nlogb (a)− ), for some constant  > 0
0
Then: limx→a fg(x)
(x)
= limx→a fg0 (x)
(x) then T (n) = θ(nlogb (a) )
2. f (n) = θ(nlogb (a) ), then T (n) = θ(nlogb (a) log(n))
Useful sums
Pn n∗(n+1) 3. f (n) = Ω(nlogb (a)+ ), for some constant  > 0 and
• i=0 i = 2 if a ∗ f (n/b) ≤ c ∗ f (n), for some constant c < 1
Pn n∗(n+1)(2n+1) and all sufficiently large n, then T (n) = θ(f (n))
• i=0 i2 = 6

•
Pn
ip ∈ θ(np+1 ) Extension of Master Theorem
i=0
Pn Assuming: f (n) = O(nlogb (a) ∗ logb (n)α ). Then:
• i=0 2i = 2(n+1) − 1
1. α < −1 → T (n) = θ(nlogb (a) )
Pn a∗(1−r n )
• i=1 a ∗ ri−1 = 1−r
2. α = −1 → T (n) = θ(nlogb (a) logb (logb (n)))
Pn x−(n+1)∗xn+1 +n∗xn+2
• i=0 i ∗ xi = (x−1)2 3. α > −1 → T (n) = θ(nlogb (a) logb (n)α+1 )
(source)
Pn
• 1
i=1 x ∈ θ(log(n))
Pn
• i=1 ln(j) = ln(n!) = n ∗ ln(n) − n + O(ln(n))