Improved Monotone Circuit Depth Upper Bound for Directed Graph Reachability

Reading time: 8 minute
...

📝 Original Info

  • Title: Improved Monotone Circuit Depth Upper Bound for Directed Graph Reachability
  • ArXiv ID: 0809.3614
  • Date: 2008-09-23
  • Authors: Researchers from original ArXiv paper

📝 Abstract

We prove that the directed graph reachability problem (transitive closure) can be solved by monotone fan-in 2 boolean circuits of depth (1/2+o(1))(log n)^2, where n is the number of nodes. This improves the previous known upper bound (1+o(1))(log n)^2. The proof is non-constructive, but we give a constructive proof of the upper bound (7/8+o(1))(log n)^2.

💡 Deep Analysis

Deep Dive into Improved Monotone Circuit Depth Upper Bound for Directed Graph Reachability.

We prove that the directed graph reachability problem (transitive closure) can be solved by monotone fan-in 2 boolean circuits of depth (1/2+o(1))(log n)^2, where n is the number of nodes. This improves the previous known upper bound (1+o(1))(log n)^2. The proof is non-constructive, but we give a constructive proof of the upper bound (7/8+o(1))(log n)^2.

📄 Full Content

arXiv:0809.3614v1 [cs.CC] 22 Sep 2008 Impro v ed Monotone Cir uit Depth Upp er Bound for Dire ted Graph Rea habilit y S. A. V olk o v (Mos o w, 2008) Abstra t W e pro v e that the dire ted graph rea habilit y problem an b e solv ed b y monotone fan-in 2 b o olean ir uits of depth (1/2 + o(1))(log2 n)2 , where n is the n um b er of no des. This impro v es the previous kno wn upp er b ound (1+o(1))(log2 n)2 . The pro of is non- onstru tiv e, but w e giv e a onstru tiv e pro of of the upp er b ound (7/8 + o(1))(log2 n)2 . Denitions By [x] denote the o or of x . F or real-v alue fun tions f, g w e will write f ≍g , if f = O(g) , and g = O(f) ; w e will write f ∼g , if lim(f/g) = 1 . By G(g11, g12, . . . , g1n, . . . , gn1, gn2, . . . , gnn) denote the dire ted graph su h that the set of v erti es of this graph is {1, . . . , n} , and the adja en y matrix of this graph is {gij}. By V (G) denote the set of v erti es of the graph G . Similarly , b y E(G) denote the set of edges of the graph G . W e sa y that the path p has length l , if p has l edges. The graph G′ is alled l - losur e of the graph G , if V (G′) = V (G) , and E(G′) = {(i, j) : G has a path from i to j with length ≤l }. Supp ose G is a graph, S ⊆V (G) ; b y GS denote the graph su h that V (GS) = S , E(GS) = {(i, j) ∈E(G) : i, j ∈S} . In the follo wing denitions G is an abbreviated notation for G(g11, . . . , gnn) . Put Reachn(g11, . . . , gnn) = ( 1, if the graph G has a path from 1 to n, 0 in the other ase, Reachn,l(g11, . . . , gnn) =      1, if the graph G has a path from 1 to n with length l , 0 in the other ase, Reach p n,l(g11, . . . , gnn) =          1, if the graph G has a path from 1 to n with length ≤l , 0, if there is no path from 1 to n in the graph G , undened in the other ases. 1 In this pap er w e onsider b o olean monotone ir uits with fan-in 2 , i.e., ir uits o v er the basis {x&y, x ∨y}. By d(Σ) denote the depth of the ir uit Σ. The ordered family of sets (S1, . . . , Sm) is alled an (n, m, s, l, d) -family , where n , m , s , l , d are natural n um b ers, if the follo wing onditions hold: 1. | Sm i=1 Si| ≤n ; 2. for an y i w e ha v e |Si| ≤s ; 3. F or an y A ⊆{1, . . . , m} su h that |A| ≥md l , w e ha v e | S i∈A Si| ≥n −d + 1 . Order is needed for al ulation on v enien e only (in the pro of of statemen t 4). Supp ose M = (S1, . . . , Sm) is a family of sets; b y S M denote the set | Sm i=1 Si| . Pro of of the upp er b ound Statemen t 1. Ther e exists a family of monotone ir uits Σn,l su h that for any n, l , Σn,l r e alizes Reachn,l, and d(Σn,l) = log2 n · log2 l + O(log n). Äîêàçàòå ëüñòâî. By rep eated matrix squaring. Statemen t 2. L et M b e an (n, m, s, l, d) -family, A0, . . . , Al′ ( l′ ≤l ) b e dier ent elements of S M ; then ther e exists a set S ∈M and numb ers k ≤l d , 0 = i0 < i1 < . . . < ik = l′ su h that Ai1, . . . , Aik−1 ∈S , and for any j ( 0 ≤j < k ) we have ij+1 −ij ≤2d . Pr o of. Without loss of generalit y w e an assume that l′ ≥2d + 1 . Supp ose k = l′ d  , Bi = {Aid, Aid+1, . . . , Aid+d−1} , 1 ≤i ≤k −1 . W e no w pro v e that there exists S ∈M su h that for an y i ( 1 ≤i ≤k −1 ) w e ha v e S ∩Bi ̸= ∅ . Assume the on v erse. Then for an y set Bi there exists an S ∈M su h that S ∩Bi = ∅ . Then there exists a set Bi su h that there exists at least m k−1 ≥md l sets S ∈M su h that S ∩Bi = ∅ . Sin e |Bi| = d , w e see that the ardinalit y of the union of these sets S do es not ex eed n −d . This on tradi ts the denition of an (n, m, s, l, d) -family . T ak e the set S ∈M su h that for an y Bj w e ha v e S ∩Bj ̸= ∅ . Also, tak e i1, . . . , ik−1 su h that Aij ∈Bj ∩S ( 1 ≤j ≤k −1 ). Put i0 = 0 , ik = l′ . Ob viously , the n um b ers i0, . . . , ik satisy the onditions of the statemen t. Statemen t 3. L et M b e an (n, m, s, l, d) -family; supp ose a monotone ir uit Σ′ r e alizes Reach p s+2,[ l d] ; then ther e exists a monotone ir uit Σ su h that Σ r e alizes Reach p n,l , and d(Σ) = log2 m + log2 n · log2 d + d(Σ′) + f(n) for some xe d fun tion f(n) = O(log n) . 2 Pr o of. Without loss of generalit y , w e an assume that all sets of the family M are subsets of {1, . . . , n} . A dding 1 and n to ea h set of M , w e obtain the (n, m, s + 2, l, d) -family M′ . No w w e des rib e a onstru tion of the ir uit Σ . Let G b e an input graph. Supp ose G′ is the 2d - losure of the graph G . F rom statemen t 1 it follo ws that the adja en y matrix of G′ an b e obtained b y a monotone ir uit of depth log2 n · log2 d + O(log n) . F or ea h set S ∈M′ w e onstru t a blo

k ΣS whi h is a lone of the ir uit Σ′ ; it tak es the adja en y matrix of G′ S to it’s input ( ΣS determines the existen e of a needed path in G′ S from the v ertex 1 to the v ertex n ); if |V (G′ S)| < s , then the input matrix of ΣS is expanded b y zeros. The disjun tion of all outputs of the ir uits ΣS , S ∈M′ is de lared to b e the output of Σ . It’s lear that d(Σ) satises the

…(Full text truncated)…

📸 Image Gallery

cover.png

Reference

This content is AI-processed based on ArXiv data.

Start searching

Enter keywords to search articles

↑↓
ESC
⌘K Shortcut