NP-hardness of Minimum Circuit Size Problem for OR-AND-MOD Circuits Shuichi Hirahara (The University of Tokyo) Igor C. Oliveira (University of Oxford) Rahul Santhanam (University of Oxford) CCC 2018 @ San Diego June 22, 2018
Talk Outline 1. MCSP and Its background 2. π - MCSP for a circuit class π 3. Our Results 4. Proof Sketch
Talk Outline 1. MCSP and Its background 2. π - MCSP for a circuit class π 3. Our Results 4. Proof Sketch
Minimum Circuit Size Problem (MCSP) Output Input β’ Truth table π β 0,1 2 π’ of a Is there a circuit of size β€ π‘ function π: 0,1 π’ β 0,1 that computes π ? β’ Size parameter π‘ β β Example: Output: βYESβ π‘ = 5 π π π π π π β π π 0 0 0 0 1 1 π = 1 0 1 1 1 0
Brief History of MCSP 1950s Recognized as an important problem in the Soviet Union [Trakhtenbrot βs survey] 1970s Levin delayed publishing his work because he wanted to say something about MCSP. Masek proved NP-completeness of DNF -MCSP. 1979 2000 Kabanets and Cai revived interest, based on natural proofs. [Razborov & Rudich (1997)] Since then many papers and results appeared; however, the complexity of MCSP remains elusive.
Current Knowledge about MCSP β’ Upper bound: MCSP β ππ β’ Lower bound: β pseudorandom function generators βΉ MCSP β π β’ Big Open Question: Is MCSP NP-hard? β’ No consensus about the exact complexity of MCSP β No strong evidence against NP-completeness β’ Weak evidence: [Hirahara- Santhanam (CCCβ17)] [Allender - Hirahara 17]β¦ β No strong evidence for NP-completeness β’ Some new evidence: [Impagliazzo-Kabanets-Volkvovich (CCCβ18)] & This work
Kabanets-Cai Obstacle: Why so difficult? β’ Suppose that we want to construct a reduction from SAT to MCSP . β¦ (π, π‘) π β SAT CircSize π β€ π‘ β¦ (π, π‘) CircSize π > π‘ π β SAT Need to prove a circuit lower bound! β’ Natural reduction techniques would imply E β SIZE π π 1 . [Kabanets-Cai (2000)]
Talk Outline 1. MCSP and Its background 2. π - MCSP for a circuit class π 3. Our Results 4. Proof Sketch
π - MCSP for a circuit class π Output Input β’ Truth table π β 0,1 2 π’ of a Is there a π -circuit of size β€ π‘ function π: 0,1 π’ β 0,1 that computes π ? β’ Size parameter π‘ β β Theorem [Masek (1978 or 79, unpublished)] DNFβMCSP is NP-hard.
DNF - MCSP Output Input β’ Truth table π β 0,1 2 π’ of a Is there a DNF formula of size function π: 0,1 π’ β 0,1 β€ π‘ that computes π ? β’ Size parameter π‘ β β Depth: 2 β¨ Size: 3 Example of DNFs: β§ β§ β§ β‘ Β¬π¦ 1 β§ π¦ 2 β¨ π¦ 2 β§ Β¬π¦ 3 β¨ Β¬π¦ 2 Β¬π¦ 1 π¦ 2 π¦ 2 Β¬π¦ 3 Β¬π¦ 2 The size of DNF β #(clauses)
π - MCSP for π β DNF β’ Beyond DNFs , no NP -hardness was proved since the work of Masek (1979). β’ To quote Allender, Hellerstein, McCabe, Pitassi, and Saks (2008): βThus an important open question is to resolve the NP- hardness of β¦ function minimization results above for classes that are stronger than DNF.β
Known results about π - MCSP More expressive MCSP Hardness based on cryptography β― e.g.) Blum integer 0 - MCSP for large π factorization [AHMPS08] AC π β― No hardness result Depth3- MCSP Known to be NP-hard DNF - MCSP [Masek (1979)] Remark: The complexity is not necessarily monotone increasing or decreasing.
Talk Outline 1. MCSP and Its background 2. π - MCSP for a circuit class π 3. Our Results 4. Proof Sketch
Our Results β’ The first NP-hardness result for π -MCSP for a class π β DNF Theorem (Main Result) DNF β XOR βMCSP is NP-hard under polynomial-time many-one reductions. β’ Our proof techniques extend to: β’ DNF β MOD π - MCSPβ² is NP-hard for any π β₯ 2 , but the input is a truth table of an π -valued function π: β€/πβ€ π’ β 0,1 .
DNF β XOR circuits ( 2 Ξ© π circuit lower bound is known) [Cohen & Shinkar (2016)] β¨ Example Depth 3 β§ β§ β§ Size 3 1 st layer: an OR gate 2 nd layer: AND gates β β β β 3 rd layer: XOR gates π¦ 2 Β¬π¦ 1 Β¬π¦ 3 Β¬π¦ 2 The size of DNF β XOR circuits β The number of AND gates β’ This is a convenient circuit size measure as advocated by Cohen & Shinkar (2016). 1. Nice combinatorial meaning 2. W.l.o.g., # XOR gates β€ π β #(AND gates) β’ Our proof techniques extend to the number of all the gates in a DNF β XOR formula.
DNF β XOR circuits ( 2 Ξ© π circuit lower bound is known) [Cohen & Shinkar (2016)] β¨ Example Depth 3 β§ β§ β§ Size 3 1 st layer: an OR gate 2 nd layer: AND gates β β β β 3 rd layer: XOR gates π¦ 2 Β¬π¦ 1 Β¬π¦ 3 Β¬π¦ 2 1 β π¦ 1 β π¦ 2 β 1 β π¦ 3 = 1 outputs 1. βΊ The subcircuit π¦ 2 β 1 β π¦ 3 = 1 β Some linear equations over GF 2 βΊ π¦ 1 , π¦ 2 , π¦ 3 β π΅ (for some affine subspace π΅ β GF 2 π )
DNF β XOR circuits ( 2 Ξ© π circuit lower bound is known) [Cohen & Shinkar (2016)] π: 0,1 π β 0,1 β¨ Example Depth 3 π΅ 3 π΅ 1 π΅ 2 β§ β§ β§ Size 3 1 st layer: an OR gate 2 nd layer: AND gates β β β β 3 rd layer: XOR gates π¦ 2 Β¬π¦ 1 Β¬π¦ 3 Β¬π¦ 2 π β1 1 = π΅ 1 βͺ π΅ 2 βͺ π΅ 3
The Important Observation The minimum DNF β XOR circuit size for computing π = The minimum number π of affine subspaces needed to cover π β1 1 : that is, βπ΅ 1 , β¦ , π΅ π : affine subspaces of 0,1 π π΅ π β π β1 1 π΅ 1 βͺ β― βͺ π΅ π = π β1 1 and
Talk Outline 1. MCSP and Its background 2. π - MCSP for a circuit class π 3. Our Results 4. Proof Sketch
β’ Our proof was inspired by a simple proof of Masekβs result given by [Allender, Hellerstein, McCabe, Pitassi, and Saks (2008)]. β’ We extend and generalize their ideas significantly.
Proof Outline Theorem (Main Result) π NP β€ π DNF β XOR βMCSP Step 1. 2-factor approx. of DNF β XOR - MCSP ZPP β€ π π -Bounded Set Cover for partial functions (NP-hard [Trevisan 2001] ) DNF β XOR - MCSP Step 2. ZPP DNF β XOR - MCSP β€ π for partial functions Derandomization using π -biased generators Step 3. [Naor & Naor (1993)]
Proof Outline Theorem (Main Result) π NP β€ π DNF β XOR βMCSP Step 1. 2-factor approx. of DNF β XOR - MCSP ZPP β€ π π -Bounded Set Cover for partial functions (NP-hard [Trevisan 2001] ) DNF β XOR - MCSP Step 2. ZPP DNF β XOR - MCSP β€ π for partial functions Derandomization using π -biased generators Step 3. [Naor & Naor (1993)]
The Set Cover Problem A universe π and a collection of sets π― β 2 π Input: Output: The minimum π such that π β π― and Ϊ π·βπ π· = π Example: π = { β¦ } , π― = { β¦ }
The Set Cover Problem A universe π and a collection of sets π― β 2 π Input: Output: The minimum π such that π β π― and Ϊ π·βπ π· = π Example: π = { β¦ } , π― = { β¦ } A minimum cover π :
The π -Bounded Set Cover Problem A universe π and a collection of sets π― β 2 π Input: such that π β€ π for every π β π― . Output: The minimum π such that π β π― and Ϊ π·βπ π· = π Example: π = { β¦ } , π― = { β¦ } 4-bounded, but not 3-bounded β’ We set π to be large enough [Feige (1998)] [Trevisan (2001)] so that a 2 -factor approx. is NP-hard. Approximation of 1 β π 1 ln π is NP -hard.
Proof Outline Theorem (Main Result) π NP β€ π DNF β XOR βMCSP Step 1. 2-factor approx. of DNF β XOR - MCSP ZPP β€ π π -Bounded Set Cover for partial functions (NP-hard [Trevisan 2001] ) DNF β XOR - MCSP Step 2. ZPP DNF β XOR - MCSP β€ π for partial functions Derandomization using π -biased generators Step 3. [Naor & Naor (1993)]
DNF β XOR - MCSP β for partial functions Output Input β’ Truth table of a partial Is there a circuit of size β€ π‘ function that agrees with π on inputs π: 0,1 π’ β 0,1,β from π β1 0,1 ? β’ Size parameter π‘ β β Example: π π π π π π π , π π 0 0 β 0 1 1 π = β 1 0 1 1 0
Claim DNF β XOR - MCSP 2-factor approx. of ZPP β€ π for partial functions π -Bounded Set Cover β’ Given: π = 1, β¦ , π , π― = {π 1 , β¦ , π π } Construct π: 0,1 π’ β 0,1,β for π’ = π log π β’ Goal: DNF β XOR - MCSP Set Cover π€ π βΌ 0,1 π’ (A uniformly random vector) π β π β¦ span πβπ π π€ π β 0,1 π’ π π β π― β¦ span πβπ π€ π β 0,1 π’ Cover π β π― β¦ α« πβπ
Set Cover DNF β XOR - MCSP π: 0,1 π’ β 0,1,β π = 1,2,3 , π― = {π 1 , π 2 } π 1 = 1,2 1 2 3 π 2 = 2,3 π π€ π β 1 for any π β π . β’ π π¦ β 0 for all π¦ β span π€ 1 , π€ 2 βͺ span π€ 2 , π€ 3 . β’ β’ π π§ β β for any other vector π§ β 0,1 π’ . β’ The minimum DNF β XOR circuit size for computing π The minimum number of affine subspaces π΅ β π β1 1,β = needed to cover π β1 1 = π€ 1 , π€ 2 , π€ 3 .
Recommend
More recommend