SDPs: basic method Algorithmic Spencers Result Lovett-Meka result Lower bounds via SDP duality (Matousek) 2/30 Material Classic: Geometric Discrepancy by J. Matousek
Papers: Bansal. Constructive algorithms for discrepancy minimization, FOCS 2010 Matousek. The determinant lower bound is almost tight Lovett, Meka. Discrepancy minimization by walking on the edges
Survey with fewer technical details: Bansal. 3/30 4/30 Discrepancy: What is it? Study of gaps in approximating the continuous by the discrete.
Original motivation: Numerical Integration/ Sampling
Problem: How well can you approximate a region by discrete points Discrepancy: Max over intervals I |(# points in I) (length of I)| 5/30 Discrepancy: What is it? Study of gaps in approximating the continuous by the discrete.
Problem: How uniformly can you distribute points in a grid.
Uniform : For every axis-parallel rectangle R | (# points in R) - (Area of R) | should be low. n 1/2 n 1/2 Discrepancy: Max over rectangles R |(# points in R) (Area of R)| 6/30 Distributing points in a grid Problem: How uniformly can you distribute points in a grid.
Uniform : For every axis-parallel rectangle R | (# points in R) - (Area of R) | should be low. Uniform Random Van der Corput Set n= 64 points n 1/2 discrepancy n 1/2 (loglog n) 1/2
O(log n) discrepancy! Quasi-Monte Carlo Methods With N random samples: Error \prop 1/\sqrt{n} Quasi-Monte Carlo Methods: \prop Disc/n
Can discrepancy be O(1) for 2d grid? No. \Omega(log n) [Schmidt ]
7/30 8/30 Discrepancy: Example 2 Input: n points placed arbitrarily in a grid. Color them red/blue such that each rectangle is colored as evenly as possible
Discrepancy: max over rect. R ( | # red in R - # blue in R | ) Continuous: Color each element 1/2 red and 1/2 blue (0 discrepancy)
Discrete: Random has about O(n 1/2 log 1/2 n) Can achieve O(log 2.5 n) 9/30 Combinatorial Discrepancy Universe: U= [1,,n] Subsets: S 1 ,S 2 ,,S m
Color elements red/blue so each set is colored as evenly as possible.
Find _: [n] ! {-1,+1} to Minimize |_(S)| 1 = max S | i 2 S _(i) |
If A is m \times n incidence matrix. Disc(A) = min_{x \in {-1,1}^n} |Ax|_\infty
S 1 S 2 S 3 S 4 10/30 Applications CS: Computational Geometry, Comb. Optimization, Monte-Carlo simulation, Machine learning, Complexity, Pseudo-Randomness,
Hereditary Discrepancy 11/30 Rounding Lovasz-Spencer-Vesztermgombi86 Given any matrix A, and x \in R^n can round x to \tilde{x} \in Z^n s.t. |Ax A\tilde{x}|_\infty < Herdisc(A)
Proof: Round the bits one by one. 12/30 Can we find it efficiently?
Nothing known until recently.
Thm [B10]. Can efficiently round so that Error \leq O(\sqrt{log m log n}) Herdisc(A) 13/30 More rounding approaches Bin Packing
Refined further by Rothvoss (Entropy rounding method) 14/30 Dynamic Data Structures N points in a 2-d region. Weights update over time.
Query: Given an axis-parallel rectangle R, determine the total weight on points in R. Preprocess: 1) Low query time 2) Low update time (upon weight change) 15/30 Example Line: Query = O(n) Update = 1
Query = 1 Update = O(n^2) Query = 2 Update = O(n)
Query = O(log n) Update = O(log n)
Recursively can get for 2-d.
16/30 What about other objects? Query
Circles arbitrary rectangles aligned triangle
Turns out t_q t_u \geq n^{1/2}/log^2 n ? Larsen-Green: t_q t_u \geq disc(S)^n/log^2 n 17/30 Sketch of idea
A good data structure implies D = A P
A = row sparse P = Column sparse (low query time) (low update time) 18/30 Outline again
19/30 Basic Results
20/30 21/30 Best Known Algorithm Random: Color each element i independently as x(i) = +1 or -1 with probability each.
Thm: Discrepancy = O (n log n) 1/2
Pf: For each set, expect O(n 1/2 ) discrepancy Standard tail bounds: Pr[ | i 2 S x(i) | c n 1/2 ] e -c 2
Union bound + Choose c (log n) 1/2
Analysis tight: Random actually incurs O((n log n) 1/2 ).
22/30 Better Colorings Exist! [Spencer 85]: (Six standard deviations suffice) Always exists coloring with discrepancy 6n 1/2
(In general for arbitrary m, discrepancy = O(n 1/2 log(m/n) 1/2 ) Tight: For m=n, cannot beat 0.5 n 1/2 (Hadamard Matrix, orthogonal sets)
Inherently non-constructive proof (pigeonhole principle on exponentially large universe)
Challenge: Can we find it algorithmically ? Certain algorithms do not work [Spencer]
Conjecture [Alon-Spencer]: May not be possible.
23/30 Beck Fiala Thm U = [1,,n] Sets: S 1 ,S 2 ,,S m
Suppose each element lies in at most t sets (t << n).
Beck Fiala Conjecture: O(t 1/2 ) discrepancy possible
Other results: O( t 1/2 log t log n ) [Beck] O( t 1/2 log n ) [Srinivasan] O( t 1/2 log 1/2 n ) [Banaszczyk] S 1 S 2 S 3 S 4
Non-constructive 24/30 Approximating Discrepancy Question: If a set system has low discrepancy (say << n 1/2 ) Can we find a good discrepancy coloring ?
[Charikar, Newman, Nikolov 11]: Even 0 vs. O (n 1/2 ) is NP-Hard
(Matousek): What if system has low Hereditary discrepancy? herdisc (U,S) = max U U disc (U, S |U )
Robust measure of discrepancy (often same as discrepancy) Widely used: TU set systems, Geomety, S 1
S 2
S 1
S 2
1 2 n 1 2 n 25/30 Our Results Thm 1: Can get Spencers bound constructively. That is, O(n 1/2 ) discrepancy for m=n sets.
Thm 2: If each element lies in at most t sets, get bound of O(t 1/2 log n) constructively (Srinivasans bound)
Thm 3: For any set system, can find Discrepancy O(log (mn)) Hereditary discrepancy. Other Problems: Constructive bounds (matching current best) k-permutation problem [Spencer, Srinivasan,Tetali] Geometric problems , 26/30 Relaxations: LPs and SDPs Not clear how to use.
Linear Program is useless. Can color each element red and blue. Discrepancy of each set = 0!
SDPs (LP on v i v j , cannot control dimension of vs)
| i 2 S v i | 2 n 8 S |v i | 2 = 1 Intended solution v i = (+1,0,,0) or (-1,0,,0). Trivially feasible: v i = e i (all v i s orthogonal) Yet, SDPs will be a major tool. 27/30 Punch line SDP very helpful if tighter bounds needed for some sets.
| i 2 S v i | 2 2 n | i 2 S v i | 2 n/log n |v i | 2 1
Not apriori clear why one can do this. Entropy Method.
Algorithm will construct coloring over time and use several SDPs in the process.
Tighter bound for S 28/30 Talk Outline Introduction
The Method Low Hereditary discrepancy -> Good coloring
Additional Ideas Spencers O(n 1/2 ) bound
Partial Coloring Method 29/30 A Question 30/30 -n n Slight improvement Can be improved to O(\sqrt{n})/2^n
If you pick a random {-1,1} coloring s w.p. say >= |a \cdot s| \leq c \sqrt{n}
2^{n-1} colorings s, with |a\cdot s| \leq c \sqrt{n}
31/30 Algorithmically Easy: 1/poly(n) (How?) Answer: Pick any poly(n) colorings.
[Karmarkar-Karp81]: \approx 1/n^log n
Huge gap: Major open question
Remark: {-1,+1} not enough. Really need color 0 also. E.g. a_1 = 1, a_2==a_n = 1/(2n)
32/30 Yet another enhancement There is a {-1,0,1} coloring with at least n/2 {-1,1}s s.t. \sum_i a_i s_i \leq n/2^{n/5}
Make buckets of size 2n/2^{n/5} At least 2^{4n/5} sums fall in same bucket
Claim: Some two s and s in same bucket and differ in at least n/2 coordinates
Again consider s = (s-s)/2 33/30 Proof of Claim Claim: Any set of 2^{4n/5} vertices of the boolean cube has
[Kleitman66] Isoperimetry for cube. Hamming ball B(v,r) has the smallest diameter for a given number of vertices.
|B(v,n/4)| < 2^{4n/5} 34/30 Spencers proof
35/30 36/30 Our Approach 37/30 Algorithm (at high level) Cube: {-1,+1} n Analysis: Few steps to reach a vertex (walk has high variance) Disc(S i ) does a random walk (with low variance) start finish Algorithm: Sticky random walk Each step generated by rounding a suitable SDP Move in various dimensions correlated, e.g. o t 1 + o t 2 0 Each dimension: An Element Each vertex: A Coloring 38/30 An SDP Hereditary disc. ) the following SDP is feasible SDP: Low discrepancy: | i 2 S j v i | 2
2
|v i | 2 = 1 Rounding: Pick random Gaussian g = (g 1 ,g 2 ,,g n ) each coordinate g i is iid N(0,1)
For each i, consider q i = g v i Obtain v i 2 R n
39/30 Properties of Rounding Lemma: If g 2 R n is random Gaussian. For any v 2 R n , g v is distributed as N(0, |v| 2 ) Pf: N(0,a 2 ) + N(0,b 2 ) = N(0,a 2 +b 2 ) g v = i v(i) g i N(0, i v(i) 2 )
1. Each q i N(0,1) 2. For each set S,
i 2 S q i = g ( i2 S v i ) N(0, 2 ) (std deviation ) SDP: |v i | 2 = 1 | i2
S v i | 2
2
Recall: q i = g v i qs mimics a low discrepancy coloring (but is not {-1,+1}) 40/30 Algorithm Overview Construct coloring iteratively. Initially: Start with coloring x 0 = (0,0,0, ,0) at t = 0. At Time t: Update coloring as x t = x t-1 + (q t 1 ,,q t n ) ( tiny: 1/n suffices)
x(i) x t (i) = (q 1 i + q 2 i + + q t i )
Color of element i: Does random walk over time with step size N(0,1) Fixed if reaches -1 or +1. time -1 +1 Set S: x t (S) = i 2 S x t (i) does a random walk w/ step N(0, 2 )
41/30 Analysis
Consider time T = O(1/ 2 )
Claim 1: With prob. , at least n/2 elements reach -1 or +1. Pf: Each element doing random walk with size . Recall: Random walk with step 1, is O(t 1/2 ) away in t steps.
A Trouble: Various element updates are correlated
Consider basic walk x(t+1) = x(t) 1 with prob Define Energy u(t) = x(t) 2
E[u(t+1)] = (x(t)+1) 2 + (x(t)-1) 2 = x(t) 2 + 1 = u(t)+1 Expected energy = n at t= n.
Claim 2: Each set has O() discrepancy in expectation. Pf: For each S, x t (S) doing random walk with step size 42/30 Analysis
Consider time T = O(1/ 2 ) Claim 1: With prob. , at least n/2 variables reach -1 or +1.
) Everything colored in O(log n) rounds.
Claim 2: Each set has O() discrepancy in expectation per round.
) Expected discrepancy of a set at end = O( log n)
Thm: Obtain a coloring with discrepancy O( log (mn)) Pf: By Chernoff, Prob. that disc(S) >= 2 Expectation + O( log m) = O( log (mn)) is tiny (poly(1/m)). 43/30 Recap At each step of walk, formulate SDP on unfixed variables. Use some (existential) property to argue SDP is feasible Rounding SDP solution -> Step of walk
Properties of walk: High Variance -> Quick convergence Low variance for discrepancy on sets -> Low discrepancy 44/30 Refinements Spencers six std deviations result: Goal: Obtain O(n 1/2 ) discrepancy for any set system on m = O(n) sets.
Random coloring has n 1/2 (log n) 1/2 discrepancy
Previous approach seems useless: Expected discrepancy for a set O(n 1/2 ), but some random walks will deviate by up to (log n) 1/2 factor
Need an additional idea to prevent this. 45/30 Spencers O(n 1/2 ) result
Partial Coloring Lemma: For any system with m sets, there exists a coloring on n/2 elements with discrepancy O(n 1/2 log 1/2 (2m/n)) [For m=n, disc = O(n 1/2 )]
Algorithm for total coloring: Repeatedly apply partial coloring lemma Total discrepancy O( n 1/2 log 1/2 2 ) [Phase 1] + O( (n/2) 1/2 log 1/2 4 ) [Phase 2] + O((n/4) 1/2 log 1/2 8 ) [Phase 3] + = O(n 1/2 ) 46/30 Proving Partial Coloring Lemma Beautiful Counting argument (entropy method + pigeonhole) Idea: Too many colorings (2 n ), but few discrepancy profiles
Key Lemma: There exist k=2 4n/5 colorings X 1 ,,X k such that every two X i , X j are similar for every set S 1 ,,S n .
Some X 1 ,X 2 differ on n/2 positions Consider X = (X 1 X 2 )/2
Pf: X(S) = (X 1 (S) X 2 (S))/2 2 [-10 n 1/2 , 10 n 1/2 ] X 1 = ( 1,-1, 1 , ,1,-1,-1) X 2 = (-1,-1,-1, ,1, 1, 1) X = ( 1, 0, 1 , ,0,-1,-1) 47/30 A useful generalization There exists a partial coloring with non-uniform discrepancy bound A S for set S
Even if A S = O( n 1/2 ) in some average sense 48/30 An SDP Suppose there exists partial coloring X: 1. On n/2 elements 2. Each set S has |X(S)| A S
SDP: Low discrepancy: | i 2 S j v i | 2 A S 2
Many colors: i |v i | 2 n/2 |v i | 2 1 Pick random Gaussian g = (g 1 ,g 2 ,,g n ) each coordinate g i is iid N(0,1)
For each i, consider q i = g v i Obtain v i 2 R n
49/30 Algorithm Initially write SDP with A S = c n 1/2
Each set S does random walk and expects to reach discrepancy of O(A S ) = O(n 1/2 )
Some sets will become problematic. Reduce their A S on the fly. Not many problematic sets, and entropy penalty low.
0 20n 1/2 30n 1/2 35n 1/2 Danger 1 Danger 2 Danger 3 50/30 Concluding Remarks Construct coloring over time by solving sequence of SDPs (guided by existence results)
Works quite generally
Can be derandomized [Bansal-Spencer] (use entropy method itself for derandomizing + usual tech.) E.g. Deterministic six standard deviations can be viewed as a way to derandomize something stronger than Chernoff bounds. 51/30 Thank You! 52/30
53/30
54/30 Rest of the talk 1. How to generate q i with required properties. 2. How to update A S over time. Show n 1/2 (log log log n) 1/2 bound. 55/30 Why so few algorithms? Often algorithms rely on continuous relaxations. Linear Program is useless. Can color each element red and blue.
Improved results of Spencer, Beck, Srinivasan, based on clever counting (entropy method). Pigeonhole Principle on exponentially large systems (seems inherently non-constructive)
56/30 Partial Coloring Lemma Suppose we have discrepancy bound A S for set S.
Consider 2 n possible colorings Signature of a coloring X: (b(S 1 ), b(S 2 ),, b(S m ))
Want partial coloring with signature (0,0,0,,0) 57/30 Progress Condition Energy increases at each step: E(t) = \sum_i x_i(t)^2 Initially energy =0, can be at most n.
Expected value of E(t) = E(t-1) + \sum_i \gamma_i(t)^2 Markovs inequality. 58/30 Missing Steps 1. How to generate the \eta_i 2. How to update \Delta_S over time 59/30 Partial Coloring If exist two colorings X 1 ,X 2
1. Same signature (b 1 ,b 2 ,,b m ) 2. Differ in at least n/2 positions.
Consider X = (X 1 X 2 )/2 1. -1 or 1 on at least n/2 positions, i.e. partial coloring 2. Has signature (0,0,0,,0) X(S) = (X 1 (S) X 2 (S)) / 2, so |X(S)| A S for all S.
Can show that there are 2 4n/5 colorings with same signature. So, some two will differ on > n/2 positions. (Pigeon Hole) X 1 = (1,-1, 1 , , 1,-1,-1) X 2 = (-1,-1,-1, , 1,1, 1) 60/30
61/30 Spencers O(n 1/2 ) result
Partial Coloring Lemma: For any system with m sets, there exists a coloring on n/2 elements with discrepancy O(n 1/2 log 1/2 (2m/n)) [For m=n, disc = O(n 1/2 )]
Algorithm for total coloring: Repeatedly apply partial coloring lemma Total discrepancy O( n 1/2 log 1/2 2 ) [Phase 1] + O( (n/2) 1/2 log 1/2 4 ) [Phase 2] + O((n/4) 1/2 log 1/2 8 ) [Phase 3] + = O(n 1/2 ) Let us prove the lemma for m = n 62/30 Proving Partial Coloring Lemma Pf: Associate with coloring X, signature = (b 1 ,b 2 ,,b n ) (b i = bucket in which X(S i ) lies ) Wish to show: There exist 2 4n/5 colorings with same signature
Choose X randomly: Induces distribution on signatures. Entropy () n/5 implies some signature has prob. 2 -n/5 .
Entropy ( ) i Entropy( b i ) [Subadditivity of Entropy]
b i = 0 w.p. 1- 2 e -50 , = 1 w.p. e -50
= 2 w.p. e -450
. -10 n 1/2 -30 n 1/2 10 n 1/2 30 n 1/2 0 2 1 -1 -2 Ent(b 1 ) 1/5 63/30 A useful generalization Partial coloring with non-uniform discrepancy A S for set S A S 3A S A S 3A S 5A S 0 1 2 -1 -2 For each set S, consider the bucketing Suffices to have s Ent (b s ) n/5
Or, if A S = s n 1/2 , then s g( s ) n/5
g() e - 2 /2 > 1 ln(1/) < 1 Bucket of n 1/2 /100 has penalty ln(100) 64/30 Recap Partial Coloring: A S 10 n 1/2 gives low entropy ) 2 4n/5 colorings exist with same signature. ) some X 1 ,X 2 with large hamming distance. (X 1 X2) /2 gives the desired partial coloring.
Trouble: 2 4n/5 /2 n is an exponentially small fraction. Only if we could find the partial coloring efficiently