Page 1
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Derandomizing the Isolation Lemma and Lower
Bounds for Circuit Size
V. Arvind and Partha MukhopadhyayThe Institute of Mathematical Sciences
India
27th August 2008
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 2
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
1 Introduction
2 Formulation of an Isolation Lemma
3 Automata Theory
4 Noncommutative Polynomial Identity Testing
5 Black-box derandomization
6 Summary
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 3
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Isolation Lemma (Mulmuley-Vazirani-Vazirani 1987)
U be a set (universe) of size n and F ⊆ 2U be any family ofsubsets of U.
Let w : U → Z+ be a weight function.
For T ⊆ U, define its weight w(T ) as w(T ) =∑
u∈T w(u).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 4
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Isolation Lemma (Mulmuley-Vazirani-Vazirani 1987)
U be a set (universe) of size n and F ⊆ 2U be any family ofsubsets of U.
Let w : U → Z+ be a weight function.
For T ⊆ U, define its weight w(T ) as w(T ) =∑
u∈T w(u).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 5
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Isolation Lemma (Mulmuley-Vazirani-Vazirani 1987)
U be a set (universe) of size n and F ⊆ 2U be any family ofsubsets of U.
Let w : U → Z+ be a weight function.
For T ⊆ U, define its weight w(T ) as w(T ) =∑
u∈T w(u).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 6
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Isolation Lemma
Let w be any random weight assignment w : U → [2n].
Isolation Lemma guarantees that with high probability (atleast 1/2) there will be a unique minimum weight set in F .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 7
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Isolation Lemma
Let w be any random weight assignment w : U → [2n].
Isolation Lemma guarantees that with high probability (atleast 1/2) there will be a unique minimum weight set in F .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 8
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Important applications of Isolation Lemma
Randomized NC algorithm for computing maximumcardinality matchings for general graphs.(Mulmuley-Vazirani-Vazirani 1987)
NL ⊂ UL/poly (Klaus Reinhardt and Eric Allender 2000).
SAT is many-one reducible via randomized reductions toUSAT.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 9
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Important applications of Isolation Lemma
Randomized NC algorithm for computing maximumcardinality matchings for general graphs.(Mulmuley-Vazirani-Vazirani 1987)
NL ⊂ UL/poly (Klaus Reinhardt and Eric Allender 2000).
SAT is many-one reducible via randomized reductions toUSAT.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 10
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Important applications of Isolation Lemma
Randomized NC algorithm for computing maximumcardinality matchings for general graphs.(Mulmuley-Vazirani-Vazirani 1987)
NL ⊂ UL/poly (Klaus Reinhardt and Eric Allender 2000).
SAT is many-one reducible via randomized reductions toUSAT.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 11
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Two outstanding open problems in complexity theory
Is the matching problem in in deterministic NC ?
Is NL ⊆ UL ?
Both the problems will be solved if Isolation Lemma can bederandomized.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 12
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Two outstanding open problems in complexity theory
Is the matching problem in in deterministic NC ?
Is NL ⊆ UL ?
Both the problems will be solved if Isolation Lemma can bederandomized.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 13
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Derandomizing Isolation Lemma
In all well known applications of Isolation Lemma number ofset system is 2nO(1)
.
So derandomization is plausible (Agrawal 2007, Barbadosworkshop on CC).
Main Question Can we derandomize some special cases of theIsolation Lemma.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 14
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Derandomizing Isolation Lemma
In all well known applications of Isolation Lemma number ofset system is 2nO(1)
.
So derandomization is plausible (Agrawal 2007, Barbadosworkshop on CC).
Main Question Can we derandomize some special cases of theIsolation Lemma.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 15
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Derandomizing Isolation Lemma
In all well known applications of Isolation Lemma number ofset system is 2nO(1)
.
So derandomization is plausible (Agrawal 2007, Barbadosworkshop on CC).
Main Question Can we derandomize some special cases of theIsolation Lemma.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 16
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Isolation Lemma - Our setting
The universe U = [n].
An n-input boolean circuit C and size(C ) = m.
Each subset S ⊆ U corresponds to its characteristic binarystring χS ∈ {0, 1}n.
n-input boolean circuit C implicitly defines the set system
FC = {S ⊆ [n] | C (χS ) = 1}.
Also, there is only exponential number of set systems.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 17
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Isolation Lemma - Our setting
The universe U = [n].
An n-input boolean circuit C and size(C ) = m.
Each subset S ⊆ U corresponds to its characteristic binarystring χS ∈ {0, 1}n.
n-input boolean circuit C implicitly defines the set system
FC = {S ⊆ [n] | C (χS ) = 1}.
Also, there is only exponential number of set systems.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 18
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Isolation Lemma - Our setting
The universe U = [n].
An n-input boolean circuit C and size(C ) = m.
Each subset S ⊆ U corresponds to its characteristic binarystring χS ∈ {0, 1}n.
n-input boolean circuit C implicitly defines the set system
FC = {S ⊆ [n] | C (χS ) = 1}.
Also, there is only exponential number of set systems.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 19
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Isolation Lemma - Our setting
The universe U = [n].
An n-input boolean circuit C and size(C ) = m.
Each subset S ⊆ U corresponds to its characteristic binarystring χS ∈ {0, 1}n.
n-input boolean circuit C implicitly defines the set system
FC = {S ⊆ [n] | C (χS ) = 1}.
Also, there is only exponential number of set systems.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 20
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Isolation Lemma - Our setting
The universe U = [n].
An n-input boolean circuit C and size(C ) = m.
Each subset S ⊆ U corresponds to its characteristic binarystring χS ∈ {0, 1}n.
n-input boolean circuit C implicitly defines the set system
FC = {S ⊆ [n] | C (χS ) = 1}.
Also, there is only exponential number of set systems.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 21
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Our Setting
w : U → [2n] : random weight assignment.
Isolation Lemma:
Probw [ There exists a unique minimum weight set in FC ] ≥1
2.
Can we derandomize?
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 22
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Our Setting
w : U → [2n] : random weight assignment.
Isolation Lemma:
Probw [ There exists a unique minimum weight set in FC ] ≥1
2.
Can we derandomize?
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 23
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Our Setting
w : U → [2n] : random weight assignment.
Isolation Lemma:
Probw [ There exists a unique minimum weight set in FC ] ≥1
2.
Can we derandomize?
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 24
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
A non black-box derandomization Hypothesis
C is an n-input boolean circuit.
A deterministic algorithm A1 takes as input (C , n).
A outputs weight functions w1,w2, · · · ,wt (wi : [n] → [2n]) :∃i , s.t wi assigns a unique minimum weight set in FC .
A1 runs in time subexponential in size(C ).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 25
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
A non black-box derandomization Hypothesis
C is an n-input boolean circuit.
A deterministic algorithm A1 takes as input (C , n).
A outputs weight functions w1,w2, · · · ,wt (wi : [n] → [2n]) :∃i , s.t wi assigns a unique minimum weight set in FC .
A1 runs in time subexponential in size(C ).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 26
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
A non black-box derandomization Hypothesis
C is an n-input boolean circuit.
A deterministic algorithm A1 takes as input (C , n).
A outputs weight functions w1,w2, · · · ,wt (wi : [n] → [2n]) :∃i , s.t wi assigns a unique minimum weight set in FC .
A1 runs in time subexponential in size(C ).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 27
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
A non black-box derandomization Hypothesis
C is an n-input boolean circuit.
A deterministic algorithm A1 takes as input (C , n).
A outputs weight functions w1,w2, · · · ,wt (wi : [n] → [2n]) :∃i , s.t wi assigns a unique minimum weight set in FC .
A1 runs in time subexponential in size(C ).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 28
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Black-box derandomization Hypothesis
A2 takes (m, n) in unary.
A outputs weight functions w1,w2, · · · ,wt (wi : [n] → [2n]).
For each size m boolean circuit C with n inputs: ∃i , s.t wi
assigns a unique minimum weight set in FC .
A2 runs in time polynomial in m.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 29
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Black-box derandomization Hypothesis
A2 takes (m, n) in unary.
A outputs weight functions w1,w2, · · · ,wt (wi : [n] → [2n]).
For each size m boolean circuit C with n inputs: ∃i , s.t wi
assigns a unique minimum weight set in FC .
A2 runs in time polynomial in m.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 30
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Black-box derandomization Hypothesis
A2 takes (m, n) in unary.
A outputs weight functions w1,w2, · · · ,wt (wi : [n] → [2n]).
For each size m boolean circuit C with n inputs: ∃i , s.t wi
assigns a unique minimum weight set in FC .
A2 runs in time polynomial in m.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 31
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Black-box derandomization Hypothesis
A2 takes (m, n) in unary.
A outputs weight functions w1,w2, · · · ,wt (wi : [n] → [2n]).
For each size m boolean circuit C with n inputs: ∃i , s.t wi
assigns a unique minimum weight set in FC .
A2 runs in time polynomial in m.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 32
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Derandomization Consequences (results)
Non black-box derandomization ⇒ either NEXP 6⊂ P/poly orPerm does not have polynomial size noncommutative
arithmetic circuits.
Black-box derandomization ⇒ an explicit multilinearpolynomial fn(x1, x2, · · · , xn) ∈ F[x1, x2, · · · , xn] (incommuting variables) does not have commutative arithmeticcircuits of size 2o(n).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 33
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Derandomization Consequences (results)
Non black-box derandomization ⇒ either NEXP 6⊂ P/poly orPerm does not have polynomial size noncommutative
arithmetic circuits.
Black-box derandomization ⇒ an explicit multilinearpolynomial fn(x1, x2, · · · , xn) ∈ F[x1, x2, · · · , xn] (incommuting variables) does not have commutative arithmeticcircuits of size 2o(n).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 34
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Non black-box derandomization : proof idea
Using Isolation Lemma, design a randomized polynomial-timeidentity testing algorithm (PIT) for small degreenoncommutative circuits.
Derandomize the algorithm (subexponential time) usingHypothesis 1.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 35
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Non black-box derandomization : proof idea
Using Isolation Lemma, design a randomized polynomial-timeidentity testing algorithm (PIT) for small degreenoncommutative circuits.
Derandomize the algorithm (subexponential time) usingHypothesis 1.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 36
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Idea behind the proof cont’d.
Noncommutative version of Impagliazzo-Kabanets 2003:Derandomizing the PIT for small degree noncommutativecircuit ⇒ either NEXP 6⊂ P/poly or permanent has nopoly-size noncommutative circuit (Arvind, Mukhopadhyay andSrinivasan 2008).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 37
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Noncommutative PIT
A noncommutative arithmetic circuit C computes apolynomial in F{x1, x2, · · · , xn} (xixj 6= xjxi ) using + and ×gate.
(Bogdanov and Wee’05) Randomized poly-time PIT fornoncommutative circuits of small degree (based on classictheorem of Amitsur and Levitzki 1950).
New algorithm is based on Isolation Lemma and AutomataTheory.
Recently, using automata theory a deterministic PIT algorithmfor noncommutative circuit computing sparse polynomial isgiven (Arvind, Mukhopadhyay and Srinivasan 2008).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 38
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Noncommutative PIT
A noncommutative arithmetic circuit C computes apolynomial in F{x1, x2, · · · , xn} (xixj 6= xjxi ) using + and ×gate.
(Bogdanov and Wee’05) Randomized poly-time PIT fornoncommutative circuits of small degree (based on classictheorem of Amitsur and Levitzki 1950).
New algorithm is based on Isolation Lemma and AutomataTheory.
Recently, using automata theory a deterministic PIT algorithmfor noncommutative circuit computing sparse polynomial isgiven (Arvind, Mukhopadhyay and Srinivasan 2008).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 39
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Noncommutative PIT
A noncommutative arithmetic circuit C computes apolynomial in F{x1, x2, · · · , xn} (xixj 6= xjxi ) using + and ×gate.
(Bogdanov and Wee’05) Randomized poly-time PIT fornoncommutative circuits of small degree (based on classictheorem of Amitsur and Levitzki 1950).
New algorithm is based on Isolation Lemma and AutomataTheory.
Recently, using automata theory a deterministic PIT algorithmfor noncommutative circuit computing sparse polynomial isgiven (Arvind, Mukhopadhyay and Srinivasan 2008).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 40
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Noncommutative PIT
A noncommutative arithmetic circuit C computes apolynomial in F{x1, x2, · · · , xn} (xixj 6= xjxi ) using + and ×gate.
(Bogdanov and Wee’05) Randomized poly-time PIT fornoncommutative circuits of small degree (based on classictheorem of Amitsur and Levitzki 1950).
New algorithm is based on Isolation Lemma and AutomataTheory.
Recently, using automata theory a deterministic PIT algorithmfor noncommutative circuit computing sparse polynomial isgiven (Arvind, Mukhopadhyay and Srinivasan 2008).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 41
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Some Automata Theory Background
A finite automaton A = (Q,Σ = {x1, · · · , xn}, δ, {q0}, {qf }).
(Q,Σ, δ, q0, qf )→ (alphabet, states set, transition function,initial state, final state).
For b ∈ Σ, the 0-1 matrix Mb ∈ F|Q|×|Q|:
Mb(q, q′) =
{
1 if δb(q) = q′,0 otherwise.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 42
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Some Automata Theory Background
A finite automaton A = (Q,Σ = {x1, · · · , xn}, δ, {q0}, {qf }).
(Q,Σ, δ, q0, qf )→ (alphabet, states set, transition function,initial state, final state).
For b ∈ Σ, the 0-1 matrix Mb ∈ F|Q|×|Q|:
Mb(q, q′) =
{
1 if δb(q) = q′,0 otherwise.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 43
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Some Automata Theory Background
A finite automaton A = (Q,Σ = {x1, · · · , xn}, δ, {q0}, {qf }).
(Q,Σ, δ, q0, qf )→ (alphabet, states set, transition function,initial state, final state).
For b ∈ Σ, the 0-1 matrix Mb ∈ F|Q|×|Q|:
Mb(q, q′) =
{
1 if δb(q) = q′,0 otherwise.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 44
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Some Automata Theory Background
For any w = w1w2 · · ·wk ∈ Σ∗, the matrixMw = Mw1Mw2 · · ·Mwk
.
Easy fact: Mw (q0, qf ) = 1 if and only if w is accepted by theautomaton A.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 45
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Some Automata Theory Background
For any w = w1w2 · · ·wk ∈ Σ∗, the matrixMw = Mw1Mw2 · · ·Mwk
.
Easy fact: Mw (q0, qf ) = 1 if and only if w is accepted by theautomaton A.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 46
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Run of an automaton over a noncommutative circuit
C be any given noncommutative arithmetic circuit computingf .
Output matrix MAout = C (Mx1 ,Mx2 · · · ,Mxn).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 47
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Run of an automaton over a noncommutative circuit
C be any given noncommutative arithmetic circuit computingf .
Output matrix MAout = C (Mx1 ,Mx2 · · · ,Mxn).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 48
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Crucial Observation
The output is always 0 when f ≡ 0.
If A accepts precisely one monomial (m) of f thenMA
out(q0, qf ) = c (coefficient of m in f is c).
That’s an identity test !!
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 49
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Crucial Observation
The output is always 0 when f ≡ 0.
If A accepts precisely one monomial (m) of f thenMA
out(q0, qf ) = c (coefficient of m in f is c).
That’s an identity test !!
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 50
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Crucial Observation
The output is always 0 when f ≡ 0.
If A accepts precisely one monomial (m) of f thenMA
out(q0, qf ) = c (coefficient of m in f is c).
That’s an identity test !!
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 51
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Identity Testing Algorithm based on Isolation Lemma
Input f ∈ F{x1, x2, · · · , xn} given by an arithmetic circuit C
of.
d be an upper bound on the degree of f .
[d ] = {1, 2, · · · , d} and [n] = {1, 2, · · · , n}.
The universe (for Isolation Lemma) U = [d ] × [n].
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 52
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Identity Testing Algorithm based on Isolation Lemma
Input f ∈ F{x1, x2, · · · , xn} given by an arithmetic circuit C
of.
d be an upper bound on the degree of f .
[d ] = {1, 2, · · · , d} and [n] = {1, 2, · · · , n}.
The universe (for Isolation Lemma) U = [d ] × [n].
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 53
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Identity Testing Algorithm based on Isolation Lemma
Input f ∈ F{x1, x2, · · · , xn} given by an arithmetic circuit C
of.
d be an upper bound on the degree of f .
[d ] = {1, 2, · · · , d} and [n] = {1, 2, · · · , n}.
The universe (for Isolation Lemma) U = [d ] × [n].
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 54
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Identity Testing Algorithm based on Isolation Lemma
Input f ∈ F{x1, x2, · · · , xn} given by an arithmetic circuit C
of.
d be an upper bound on the degree of f .
[d ] = {1, 2, · · · , d} and [n] = {1, 2, · · · , n}.
The universe (for Isolation Lemma) U = [d ] × [n].
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 55
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Identity Testing Algorithm
Let v = xi1xi2 · · · xit be a nonzero monomial of f .
Identify v with Sv ⊂ U :
Sv = {(1, i1), (2, i2), · · · , (t, it)}
Set system:
F = {Sv | v is a nonzero monomial in f }
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 56
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Identity Testing Algorithm
Let v = xi1xi2 · · · xit be a nonzero monomial of f .
Identify v with Sv ⊂ U :
Sv = {(1, i1), (2, i2), · · · , (t, it)}
Set system:
F = {Sv | v is a nonzero monomial in f }
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 57
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Identity Testing Algorithm
Let v = xi1xi2 · · · xit be a nonzero monomial of f .
Identify v with Sv ⊂ U :
Sv = {(1, i1), (2, i2), · · · , (t, it)}
Set system:
F = {Sv | v is a nonzero monomial in f }
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 58
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Intuition behind the Identity Testing Algorithm
Assign random weights from [2dn] to the elements of U,
(Isolation Lemma) With probability at least 1/2, there is aunique minimum weight set in F .
Goal Construct a family of small size automatons{Aw ,t}w∈[2nd2 ],t∈[d]:
Aw ,t precisely accepts all the strings (corresponding to themonomials) v of length t, such that the weight of Sv is w .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 59
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Intuition behind the Identity Testing Algorithm
Assign random weights from [2dn] to the elements of U,
(Isolation Lemma) With probability at least 1/2, there is aunique minimum weight set in F .
Goal Construct a family of small size automatons{Aw ,t}w∈[2nd2 ],t∈[d]:
Aw ,t precisely accepts all the strings (corresponding to themonomials) v of length t, such that the weight of Sv is w .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 60
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Intuition behind the Identity Testing Algorithm
Assign random weights from [2dn] to the elements of U,
(Isolation Lemma) With probability at least 1/2, there is aunique minimum weight set in F .
Goal Construct a family of small size automatons{Aw ,t}w∈[2nd2 ],t∈[d]:
Aw ,t precisely accepts all the strings (corresponding to themonomials) v of length t, such that the weight of Sv is w .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 61
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Intuition behind the Identity Testing Algorithm
Assign random weights from [2dn] to the elements of U,
(Isolation Lemma) With probability at least 1/2, there is aunique minimum weight set in F .
Goal Construct a family of small size automatons{Aw ,t}w∈[2nd2 ],t∈[d]:
Aw ,t precisely accepts all the strings (corresponding to themonomials) v of length t, such that the weight of Sv is w .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 62
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Intuition of the Identity Testing algorithm
For each A ∈ {Aw ,t} compute the run of A on C .
(Using the isolation lemma) The automata corresponding tothe minimum weight will precisely accept (isolate) only onestring (monomial).
The automata family is easy to construct.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 63
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Intuition of the Identity Testing algorithm
For each A ∈ {Aw ,t} compute the run of A on C .
(Using the isolation lemma) The automata corresponding tothe minimum weight will precisely accept (isolate) only onestring (monomial).
The automata family is easy to construct.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 64
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Intuition of the Identity Testing algorithm
For each A ∈ {Aw ,t} compute the run of A on C .
(Using the isolation lemma) The automata corresponding tothe minimum weight will precisely accept (isolate) only onestring (monomial).
The automata family is easy to construct.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 65
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Crucial Observation
C be a noncommutative arithmetic circuit of small degree andm is a given monomial.
Easy algorithm to check if m is a nonzero monomial in C .
Construct an automaton A that accepts only m and computerun on C .
Thus, a boolean circuit C (of size poly(size(C ))), FC
definesthe monomials of C .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 66
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Crucial Observation
C be a noncommutative arithmetic circuit of small degree andm is a given monomial.
Easy algorithm to check if m is a nonzero monomial in C .
Construct an automaton A that accepts only m and computerun on C .
Thus, a boolean circuit C (of size poly(size(C ))), FC
definesthe monomials of C .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 67
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Crucial Observation
C be a noncommutative arithmetic circuit of small degree andm is a given monomial.
Easy algorithm to check if m is a nonzero monomial in C .
Construct an automaton A that accepts only m and computerun on C .
Thus, a boolean circuit C (of size poly(size(C ))), FC
definesthe monomials of C .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 68
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Crucial Observation
C be a noncommutative arithmetic circuit of small degree andm is a given monomial.
Easy algorithm to check if m is a nonzero monomial in C .
Construct an automaton A that accepts only m and computerun on C .
Thus, a boolean circuit C (of size poly(size(C ))), FC
definesthe monomials of C .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 69
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Non black-box derandomization
Given noncommutative arithmetic circuit C .
Compute boolean circuit C .
A1(C , n) = {w1,w2, · · · ,wn}.
Identity testing using {wi}’s.
Run time: subexp(size(C , n)) .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 70
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Non black-box derandomization
Given noncommutative arithmetic circuit C .
Compute boolean circuit C .
A1(C , n) = {w1,w2, · · · ,wn}.
Identity testing using {wi}’s.
Run time: subexp(size(C , n)) .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 71
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Non black-box derandomization
Given noncommutative arithmetic circuit C .
Compute boolean circuit C .
A1(C , n) = {w1,w2, · · · ,wn}.
Identity testing using {wi}’s.
Run time: subexp(size(C , n)) .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 72
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Non black-box derandomization
Given noncommutative arithmetic circuit C .
Compute boolean circuit C .
A1(C , n) = {w1,w2, · · · ,wn}.
Identity testing using {wi}’s.
Run time: subexp(size(C , n)) .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 73
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Non black-box derandomization
Given noncommutative arithmetic circuit C .
Compute boolean circuit C .
A1(C , n) = {w1,w2, · · · ,wn}.
Identity testing using {wi}’s.
Run time: subexp(size(C , n)) .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 74
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Consequence of Hypothesis 2
Goal To construct an explicit multilinear polynomial f inF[x1, x2, · · · , xn] that does not have 2o(n) size arithmeticcircuit.
Define a multilinear polynomial:
f (x1, x2, · · · , xn) =∑
S⊆[n]
cS
∏
i∈S
xi ,
we need to fix cS ’s suitably.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 75
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Consequence of Hypothesis 2
Goal To construct an explicit multilinear polynomial f inF[x1, x2, · · · , xn] that does not have 2o(n) size arithmeticcircuit.
Define a multilinear polynomial:
f (x1, x2, · · · , xn) =∑
S⊆[n]
cS
∏
i∈S
xi ,
we need to fix cS ’s suitably.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 76
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Consequence of Hypothesis 2
Goal To construct an explicit multilinear polynomial f inF[x1, x2, · · · , xn] that does not have 2o(n) size arithmeticcircuit.
Define a multilinear polynomial:
f (x1, x2, · · · , xn) =∑
S⊆[n]
cS
∏
i∈S
xi ,
we need to fix cS ’s suitably.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 77
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Important Observation
Let f be a multilinear polynomial given by a circuit C andm =
∏
i∈S xi is a monomial.
A small size boolean circuit C can decide whether m is anonzero monomial in f .
Just substitute y for each xi such that i ∈ S and 0 otherwise.
C evaluates C to check whether the coefficient of themaximum degree of y is nonzero.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 78
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Important Observation
Let f be a multilinear polynomial given by a circuit C andm =
∏
i∈S xi is a monomial.
A small size boolean circuit C can decide whether m is anonzero monomial in f .
Just substitute y for each xi such that i ∈ S and 0 otherwise.
C evaluates C to check whether the coefficient of themaximum degree of y is nonzero.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 79
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Important Observation
Let f be a multilinear polynomial given by a circuit C andm =
∏
i∈S xi is a monomial.
A small size boolean circuit C can decide whether m is anonzero monomial in f .
Just substitute y for each xi such that i ∈ S and 0 otherwise.
C evaluates C to check whether the coefficient of themaximum degree of y is nonzero.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 80
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Important Observation
Let f be a multilinear polynomial given by a circuit C andm =
∏
i∈S xi is a monomial.
A small size boolean circuit C can decide whether m is anonzero monomial in f .
Just substitute y for each xi such that i ∈ S and 0 otherwise.
C evaluates C to check whether the coefficient of themaximum degree of y is nonzero.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 81
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Consequence of Hypothesis 2
Let w1,w2, · · · ,wt are the weight functions output by A2,t ≤ mc where m is the size of the boolean circuit that definesthe monomial of f .
Let wi = (wi ,1,wi ,2, · · · ,wi ,n).
Goal is to fool every weight function wi .
For all i , write down the equation
gi (y) = f (ywi,1 , ywi,2, · · · , ywi,n) = 0.
.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 82
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Consequence of Hypothesis 2
Let w1,w2, · · · ,wt are the weight functions output by A2,t ≤ mc where m is the size of the boolean circuit that definesthe monomial of f .
Let wi = (wi ,1,wi ,2, · · · ,wi ,n).
Goal is to fool every weight function wi .
For all i , write down the equation
gi (y) = f (ywi,1 , ywi,2, · · · , ywi,n) = 0.
.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 83
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Consequence of Hypothesis 2
Let w1,w2, · · · ,wt are the weight functions output by A2,t ≤ mc where m is the size of the boolean circuit that definesthe monomial of f .
Let wi = (wi ,1,wi ,2, · · · ,wi ,n).
Goal is to fool every weight function wi .
For all i , write down the equation
gi (y) = f (ywi,1 , ywi,2, · · · , ywi,n) = 0.
.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 84
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Consequence of Hypothesis 2
Let w1,w2, · · · ,wt are the weight functions output by A2,t ≤ mc where m is the size of the boolean circuit that definesthe monomial of f .
Let wi = (wi ,1,wi ,2, · · · ,wi ,n).
Goal is to fool every weight function wi .
For all i , write down the equation
gi (y) = f (ywi,1 , ywi,2, · · · , ywi,n) = 0.
.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 85
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Consequence of Hypothesis 2
The degree of gi (y) is ≤ 2n2.
Total number of linear constraints for cS ’s is at most2n2mc < 2n for m = 2o(n).
There always exists a nontrivial solution for f .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 86
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Consequence of Hypothesis 2
The degree of gi (y) is ≤ 2n2.
Total number of linear constraints for cS ’s is at most2n2mc < 2n for m = 2o(n).
There always exists a nontrivial solution for f .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 87
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Consequence of Hypothesis 2
The degree of gi (y) is ≤ 2n2.
Total number of linear constraints for cS ’s is at most2n2mc < 2n for m = 2o(n).
There always exists a nontrivial solution for f .
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 88
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Finishing the proof
Let f has a arithmetic circuit of size 2o(n),
Then a boolean circuit C of size 2o(n) defines the monomialsof f .
Then for some weight function wi there is a unique monomial∏
j∈S xj such that∑
j∈S wi ,j takes the minimum value (by theproperty of A2).
So the polynomial gi (y) 6= 0, a contradiction.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 89
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Finishing the proof
Let f has a arithmetic circuit of size 2o(n),
Then a boolean circuit C of size 2o(n) defines the monomialsof f .
Then for some weight function wi there is a unique monomial∏
j∈S xj such that∑
j∈S wi ,j takes the minimum value (by theproperty of A2).
So the polynomial gi (y) 6= 0, a contradiction.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 90
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Finishing the proof
Let f has a arithmetic circuit of size 2o(n),
Then a boolean circuit C of size 2o(n) defines the monomialsof f .
Then for some weight function wi there is a unique monomial∏
j∈S xj such that∑
j∈S wi ,j takes the minimum value (by theproperty of A2).
So the polynomial gi (y) 6= 0, a contradiction.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 91
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Finishing the proof
Let f has a arithmetic circuit of size 2o(n),
Then a boolean circuit C of size 2o(n) defines the monomialsof f .
Then for some weight function wi there is a unique monomial∏
j∈S xj such that∑
j∈S wi ,j takes the minimum value (by theproperty of A2).
So the polynomial gi (y) 6= 0, a contradiction.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 92
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Other Result
(Spielman and Klivans 2001) Randomized PIT for smalldegree (commutative) polynomial based on a more generalformulation of isolation lemma.
Observation Derandomization of the corresponding isolationlemma imply the result of Impagliazzo and Kabanets 2003.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 93
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Other Result
(Spielman and Klivans 2001) Randomized PIT for smalldegree (commutative) polynomial based on a more generalformulation of isolation lemma.
Observation Derandomization of the corresponding isolationlemma imply the result of Impagliazzo and Kabanets 2003.
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 94
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Summary
We study the connections between derandomization ofIsolation Lemma and circuit lower bounds.
We formulate versions of Isolation Lemma based on setsystem defined by boolean circuits.
A (non black-box) derandomization of above implies circuitlower bound in the noncommutative model.
A black-box derandomization yields a circuit lower bound inusual commutative model.
The derandomization of the Isolation Lemma used bySpielman-Klivans (2001) implies the result of Impagliazzo andKabanets (2003).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 95
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Summary
We study the connections between derandomization ofIsolation Lemma and circuit lower bounds.
We formulate versions of Isolation Lemma based on setsystem defined by boolean circuits.
A (non black-box) derandomization of above implies circuitlower bound in the noncommutative model.
A black-box derandomization yields a circuit lower bound inusual commutative model.
The derandomization of the Isolation Lemma used bySpielman-Klivans (2001) implies the result of Impagliazzo andKabanets (2003).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 96
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Summary
We study the connections between derandomization ofIsolation Lemma and circuit lower bounds.
We formulate versions of Isolation Lemma based on setsystem defined by boolean circuits.
A (non black-box) derandomization of above implies circuitlower bound in the noncommutative model.
A black-box derandomization yields a circuit lower bound inusual commutative model.
The derandomization of the Isolation Lemma used bySpielman-Klivans (2001) implies the result of Impagliazzo andKabanets (2003).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 97
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Summary
We study the connections between derandomization ofIsolation Lemma and circuit lower bounds.
We formulate versions of Isolation Lemma based on setsystem defined by boolean circuits.
A (non black-box) derandomization of above implies circuitlower bound in the noncommutative model.
A black-box derandomization yields a circuit lower bound inusual commutative model.
The derandomization of the Isolation Lemma used bySpielman-Klivans (2001) implies the result of Impagliazzo andKabanets (2003).
V. Arvind, Partha Mukhopadhyay Isolation Lemma
Page 98
OutlineIntroduction
Formulation of an Isolation LemmaAutomata Theory
Noncommutative Polynomial Identity TestingBlack-box derandomization
Summary
Summary
We study the connections between derandomization ofIsolation Lemma and circuit lower bounds.
We formulate versions of Isolation Lemma based on setsystem defined by boolean circuits.
A (non black-box) derandomization of above implies circuitlower bound in the noncommutative model.
A black-box derandomization yields a circuit lower bound inusual commutative model.
The derandomization of the Isolation Lemma used bySpielman-Klivans (2001) implies the result of Impagliazzo andKabanets (2003).
V. Arvind, Partha Mukhopadhyay Isolation Lemma