lecture12

# lecture12 - 1 15.083J/6.859J Integer Optimization Lecture...

This preview shows pages 1–4. Sign up to view the full content.

1 15.083J/6.859J Integer Optimization Lecture 12: Lattices III

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
2 6.4 The approximate nearest vector problem In this section, we use Algorithm 6.3 to produce a reduced basis of a lattice, and then use this reduced basis to solve approximately the nearest vector problem in L introduced in Eq. (6.2) In particular, let L be a lattice and x Q n not belonging to L . The nearest vector problem is NP -hard. Algorithm 6.4 that we present next Fnds a b ∈L such that k b x k≤ (2 n 1) 1 / 2 min {k b x k : b ∈L\{ 0 }} . (6.12) Algorithm 6.4 Approximate nearest vector algorithm Input : A reduced basis b 1 ,..., b n of a lattice L , the Gram-Schmidt orthogonalization ˜ b 1 ˜ b n and x Q n . Output : A vector b satisfying Eq. (6.12) and rational multipliers λ i , i =1 ,...,n with | λ i |≤ 1 / 2 , such that b x = n X i =1 λ i ˜ b i . Algorithm : 1. Set z n +1 = 0 and x n +1 = x . 2. For i = n,. .., 1 : (a) Compute σ 1 ,i ,...σ i,i Q such that x i +1 = i X j =1 σ j,i ˜ b j . (b) Set λ i = b σ i,i e− σ i,i ,where b σ i,i e denotes the nearest integer to σ i,i . (c) Set z i = z i +1 + b σ i,i e b i . (d) Set x i = x i +1 −b σ i,i e b i + λ i ˜ b i . 3. Return b = z 1 and λ 1 ,...,λ n . Example 6.9 We revisit Example 6.6. We have seen that the vectors b 1 = (1 , 1) 0 , b 2 =(2 , 1) 0 constitute a reduced basis of the lattice L ( b 1 , b 2 ). The cor- responding Gram-Schmidt orthogonalization is ˜ b 1 =(1 , 1) 0 and ˜ b 2 . 5 , 1 . 5). Let x =(7 . 5 , 3 . 0) 0 . We start the approximate nearest vector algorithm by setting z 3 = 0 and x 3 = x .W et h e ns e t i = 2 and compute multipliers σ 1 , 2 2 , 2 such that x = σ 1 , 2 ˜ b 1 + σ 2 , 2 ˜ b 2 . This gives σ 1 , 2 = x 0 ˜ b 1 k ˜ b 1 k 2 =5 . 25 , and σ 2 , 2 = x 0 ˜ b 2 k ˜ b 2 k 2 . 5 .
Sec. 6.4 The approximate nearest vector problem 3 This leads to λ 2 = b σ 2 , 2 e− σ 2 , 2 =1 1 . 5= 0 . 5 z 2 = z 3 + b σ 2 , 2 e b 2 = 0 + b 2 =(2 , 1) 0 x 2 = x 3 −b σ 2 , 2 e b 2 + λ 2 ˜ b 2 = x b 2 0 . 5 ˜ b 2 =(4 . 75 , 4 . 75) 0 . We continue for i = 1. We compute σ 1 , 1 such that x 2 = σ 1 , 1 ˜ b 1 , leading to σ 1 , 1 =4 . 75, λ 1 = b σ 1 , 1 σ 1 , 1 =0 . 25, and z 1 = z 2 + b σ 1 , 1 e b 1 =(7 , 4) 0 . Note that b = z 1 , 4) 0 ∈L and b x = λ 1 ˜ b 1 + λ 2 ˜ b 2 . We next show that Algorithm 6.4 correctly fnds b satisFying Eq. (6.12). Theorem 6.10 The vector b = z 1 computed by the approximate nearest vector algorithm belongs to L and satisfes k b x k≤ (2 n 1) 1 / 2 min {k b x k : b ∈L\{ 0 }} . Proof. Let b 1 ,..., b n be the reduced basis oF L , which is part oF the input oF the approximate nearest vector algorithm. Let ˜ b 1 ˜ b n denote its Gram- Schmidt orthogonalization. Solving the recursions in Steps 2c and 2d oF the approximate nearest vector algorithm, we obtain x 1 = x n +1 n X i =1 b σ i,i e b i + n X i =1 λ i ˜ b i z 1 = z n +1 + n X i =1 b σ i,i e b i . Since x 1 = x 2 σ 1 , 1 e b 1 + λ 1 ˜ b 1 = σ 1 , 1 ˜ b 1 σ 1 , 1 e b 1 + λ 1 ˜ b 1 , and ˜ b 1 = b 1 , it Follows x 1 = 0 . Moreover, since x n +1 = x , x = n X i =1 b σ i,i e b i n X i =1 λ i ˜ b i .

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

## This note was uploaded on 10/13/2010 for the course CS 15.083J taught by Professor Dimitrisbertsimas during the Spring '04 term at MIT.

### Page1 / 15

lecture12 - 1 15.083J/6.859J Integer Optimization Lecture...

This preview shows document pages 1 - 4. Sign up to view the full document.

View Full Document
Ask a homework question - tutors are online