Analysis of Boolean Functions in Lean

4.3. Random restrictions🔗

Definition4.3.1
Group: Chapter 4: DNF formulas and small-depth circuits (44)
Group member previews
Preview
Definition 4.1.1
Loading preview
Group member preview content is loaded from the rendered-fragment cache.
uses 0
Used by 7
Reverse dependency previews
Preview
Proposition 4.3.3
Loading preview
Reverse dependency preview content is loaded from the rendered-fragment cache.
L∃∀N

Definition 4.15. For \delta\in[0,1], a set J is a \delta-random subset of a finite set N if each element of N is included independently with probability \delta. A \delta-random restriction on \{-1,1\}^n is a pair (\boldsymbol J\mid\boldsymbol z) obtained by first drawing a \delta-random subset \boldsymbol J\subseteq[n] and then drawing \boldsymbol z\sim\{-1,1\}^{\overline{\boldsymbol J}} uniformly. Coordinate i is free if i\in\boldsymbol J and fixed otherwise. Equivalently, each coordinate is independently free with probability \delta and fixed to \pm1 with probability (1-\delta)/2 each.

Lean code for Definition4.3.19 declarations
  • defdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    def FABL.deltaRandomSubsetWeight (n : ) (δ : ) (J : Finset (Fin n)) : 
    def FABL.deltaRandomSubsetWeight (n : )
      (δ : ) (J : Finset (Fin n)) : 
    Probability weight of a free set under a `δ`-random subset model. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.deltaRandomSubsetWeight_nonneg (n : ) {δ : } (hδ0 : 0  δ)
      (hδ1 : δ  1) (J : Finset (Fin n)) :
      0  FABL.deltaRandomSubsetWeight n δ J
    theorem FABL.deltaRandomSubsetWeight_nonneg
      (n : ) {δ : } (hδ0 : 0  δ)
      (hδ1 : δ  1) (J : Finset (Fin n)) :
      0  FABL.deltaRandomSubsetWeight n δ J
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.sum_deltaRandomSubsetWeight (n : ) (δ : ) :
       J, FABL.deltaRandomSubsetWeight n δ J = 1
    theorem FABL.sum_deltaRandomSubsetWeight (n : )
      (δ : ) :
       J,
          FABL.deltaRandomSubsetWeight n δ J =
        1
    The `δ`-random subset weights form a probability distribution. 
  • defdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    def FABL.expectDeltaRandomSubset (n : ) (δ : ) (g : Finset (Fin n)  ) :
      
    def FABL.expectDeltaRandomSubset (n : )
      (δ : ) (g : Finset (Fin n)  ) : 
    Expectation under a `δ`-random free set. 
  • defdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    def FABL.expectRandomRestriction (n : ) (δ : )
      (g : (J : Finset (Fin n))  FABL.FixedSignCube J  ) : 
    def FABL.expectRandomRestriction (n : )
      (δ : )
      (g :
        (J : Finset (Fin n)) 
          FABL.FixedSignCube J  ) :
      
    O'Donnell, Definition 4.15: expectation over a `δ`-random restriction `(J | z)`. 
  • defdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    def FABL.coordRestrictionWeight : FABL.CoordRestriction  
    def FABL.coordRestrictionWeight :
      FABL.CoordRestriction  
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.sum_coordRestrictionWeight :  c, FABL.coordRestrictionWeight c = 1
    theorem FABL.sum_coordRestrictionWeight :
       c, FABL.coordRestrictionWeight c = 1
  • defdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    def FABL.restrictionAssignmentWeight {n : }
      (ρ : Fin n  FABL.CoordRestriction) : 
    def FABL.restrictionAssignmentWeight {n : }
      (ρ : Fin n  FABL.CoordRestriction) : 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.sum_restrictionAssignmentWeight {n : } :
       ρ, FABL.restrictionAssignmentWeight ρ = 1
    theorem FABL.sum_restrictionAssignmentWeight
      {n : } :
       ρ,
          FABL.restrictionAssignmentWeight ρ =
        1
Definition4.3.2
Group: Chapter 4: DNF formulas and small-depth circuits (44)
Group member previews
Preview
Definition 4.1.1
Loading preview
Group member preview content is loaded from the rendered-fragment cache.
uses 1
Used by 3
Reverse dependency previews
Preview
Proposition 4.3.3
Loading preview
Reverse dependency preview content is loaded from the rendered-fragment cache.
L∃∀N

Definition 4.16. Given f:\{-1,1\}^n\to\mathbb R, free coordinates I\subseteq[n], and a fixing z\in\{-1,1\}^{\overline I} of the remaining coordinates, one may identify the restricted function f_{I\mid z}:\{-1,1\}^I\to\mathbb R with its extension f_{I\mid z}:\{-1,1\}^n\to\mathbb R that holds the coordinates in \overline I fixed at z and ignores the values of those fixed coordinates as free inputs. When dealing with random restrictions this extension convention is the default.

Lean code for Definition4.3.23 declarations
  • defdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    def FABL.extendedSignRestriction.{u_1} {n : } {α : Type u_1}
      (f : FABL.SignCube n  α) (J : Finset (Fin n))
      (z : FABL.FixedSignCube J) : FABL.SignCube n  α
    def FABL.extendedSignRestriction.{u_1} {n : }
      {α : Type u_1} (f : FABL.SignCube n  α)
      (J : Finset (Fin n))
      (z : FABL.FixedSignCube J) :
      FABL.SignCube n  α
    O'Donnell, Definition 4.16: restricted function extended to the full cube. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.extendedSignRestriction_apply.{u_1} {n : } {α : Type u_1}
      (f : FABL.SignCube n  α) (J : Finset (Fin n))
      (z : FABL.FixedSignCube J) (x : FABL.SignCube n) :
      FABL.extendedSignRestriction f J z x =
        f (FABL.combineSignCube J (fun i => x i) z)
    theorem FABL.extendedSignRestriction_apply.{u_1}
      {n : } {α : Type u_1}
      (f : FABL.SignCube n  α)
      (J : Finset (Fin n))
      (z : FABL.FixedSignCube J)
      (x : FABL.SignCube n) :
      FABL.extendedSignRestriction f J z x =
        f
          (FABL.combineSignCube J
            (fun i => x i) z)
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.extendedSignRestriction_setCoordinate_of_not_mem.{u_1} {n : }
      {α : Type u_1} (f : FABL.SignCube n  α) (J : Finset (Fin n))
      (z : FABL.FixedSignCube J) (i : Fin n) (hi : i  J)
      (x : FABL.SignCube n) (b : FABL.Sign) :
      FABL.extendedSignRestriction f J z (FABL.setCoordinate x i b) =
        FABL.extendedSignRestriction f J z x
    theorem FABL.extendedSignRestriction_setCoordinate_of_not_mem.{u_1}
      {n : } {α : Type u_1}
      (f : FABL.SignCube n  α)
      (J : Finset (Fin n))
      (z : FABL.FixedSignCube J) (i : Fin n)
      (hi : i  J) (x : FABL.SignCube n)
      (b : FABL.Sign) :
      FABL.extendedSignRestriction f J z
          (FABL.setCoordinate x i b) =
        FABL.extendedSignRestriction f J z x
Proposition4.3.3
Group: Chapter 4: DNF formulas and small-depth circuits (44)
Group member previews
Preview
Definition 4.1.1
Loading preview
Group member preview content is loaded from the rendered-fragment cache.
Statement uses 3
Statement dependency previews
Preview
Corollary 3.3.7
Loading preview
Statement dependency preview content is loaded from the rendered-fragment cache.
Used by 4
Reverse dependency previews
Preview
Corollary 4.3.4
Loading preview
Reverse dependency preview content is loaded from the rendered-fragment cache.
L∃∀N

Proposition 4.17. Fix f:\{-1,1\}^n\to\mathbb R and S\subseteq[n]. If (\boldsymbol J\mid\boldsymbol z) is a \delta-random restriction on \{-1,1\}^n, then (treating restricted functions as maps \{-1,1\}^n\to\mathbb R) \mathbb E\bigl[\widehat{f_{\boldsymbol J\mid\boldsymbol z}}(S)\bigr] =\Pr[S\subseteq\boldsymbol J]\,\widehat f(S) =\delta^{|S|}\,\widehat f(S), and \mathbb E\bigl[\widehat{f_{\boldsymbol J\mid\boldsymbol z}}(S)^2\bigr] =\sum_{U\subseteq[n]}\Pr[U\cap\boldsymbol J=S]\,\widehat f(U)^2 =\sum_{U\supseteq S}\delta^{|S|}(1-\delta)^{|U\setminus S|}\,\widehat f(U)^2. For the first identity, use Corollary 3.22 and \sum_{J\supseteq S}w_\delta(J)=\delta^{|S|}. For the second, apply the free/fixed Parseval identity on each free set, reindex by the ambient frequencies satisfying U\cap J=S, and use \sum_{J:U\cap J=S}w_\delta(J)=\delta^{|S|}(1-\delta)^{|U\setminus S|} when S\subseteq U.

Lean code for Proposition4.3.313 declarations
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_fourierCoeff_empty_randomRestriction {n : }
      (f : FABL.SignCube n  ) (δ : ) :
      (FABL.expectRandomRestriction n δ fun J z =>
          FABL.restrictionFourierCoeff f J  z) =
        FABL.fourierCoeff f 
    theorem FABL.expect_fourierCoeff_empty_randomRestriction
      {n : } (f : FABL.SignCube n  )
      (δ : ) :
      (FABL.expectRandomRestriction n δ
          fun J z =>
          FABL.restrictionFourierCoeff f J 
            z) =
        FABL.fourierCoeff f 
    O'Donnell, Proposition 4.17 (empty-set first moment). 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.sum_deltaRandomSubsetWeight_supset (n : ) (δ : )
      (S : Finset (Fin n)) :
      (∑ J, if S  J then FABL.deltaRandomSubsetWeight n δ J else 0) =
        δ ^ S.card
    theorem FABL.sum_deltaRandomSubsetWeight_supset
      (n : ) (δ : ) (S : Finset (Fin n)) :
      (∑ J,
          if S  J then
            FABL.deltaRandomSubsetWeight n δ J
          else 0) =
        δ ^ S.card
    Weight of free sets containing a fixed set `S` equals `δ ^ |S|`. 
  • defdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    def FABL.ambientRestrictionFourierCoeff {n : } (f : FABL.SignCube n  )
      (J S : Finset (Fin n)) (z : FABL.FixedSignCube J) : 
    def FABL.ambientRestrictionFourierCoeff
      {n : } (f : FABL.SignCube n  )
      (J S : Finset (Fin n))
      (z : FABL.FixedSignCube J) : 
    Free-cube Fourier coefficient of a restriction, zero unless `S ⊆ J`.
    
    This is the book’s first-moment input for Proposition 4.17 under the convention that
    frequencies outside the free set do not contribute.
    
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.ambientRestrictionFourierCoeff_eq {n : } (f : FABL.SignCube n  )
      (J S : Finset (Fin n)) (z : FABL.FixedSignCube J) :
      FABL.ambientRestrictionFourierCoeff f J S z =
        if S  J then
          FABL.restrictionFourierCoeff f J (FABL.freeFrequencyPart J S) z
        else 0
    theorem FABL.ambientRestrictionFourierCoeff_eq
      {n : } (f : FABL.SignCube n  )
      (J S : Finset (Fin n))
      (z : FABL.FixedSignCube J) :
      FABL.ambientRestrictionFourierCoeff f J
          S z =
        if S  J then
          FABL.restrictionFourierCoeff f J
            (FABL.freeFrequencyPart J S) z
        else 0
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_ambientRestrictionFourierCoeff {n : }
      (f : FABL.SignCube n  ) (J S : Finset (Fin n)) :
      (Finset.univ.expect fun z =>
          FABL.ambientRestrictionFourierCoeff f J S z) =
        (if S  J then 1 else 0) * FABL.fourierCoeff f S
    theorem FABL.expect_ambientRestrictionFourierCoeff
      {n : } (f : FABL.SignCube n  )
      (J S : Finset (Fin n)) :
      (Finset.univ.expect fun z =>
          FABL.ambientRestrictionFourierCoeff
            f J S z) =
        (if S  J then 1 else 0) *
          FABL.fourierCoeff f S
    Conditional first moment: free-set Fourier recovers ambient when `S ⊆ J`. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_fourierCoeff_randomRestriction {n : }
      (f : FABL.SignCube n  ) (δ : ) (S : Finset (Fin n)) :
      (FABL.expectRandomRestriction n δ fun J z =>
          FABL.ambientRestrictionFourierCoeff f J S z) =
        δ ^ S.card * FABL.fourierCoeff f S
    theorem FABL.expect_fourierCoeff_randomRestriction
      {n : } (f : FABL.SignCube n  )
      (δ : ) (S : Finset (Fin n)) :
      (FABL.expectRandomRestriction n δ
          fun J z =>
          FABL.ambientRestrictionFourierCoeff
            f J S z) =
        δ ^ S.card * FABL.fourierCoeff f S
    O'Donnell, Proposition 4.17 (general first moment).
    
    `E[̂f_{J|z}(S)] = δ^{|S|} ̂f(S)`, with frequencies outside the free set treated as zero.
    
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.liftFreeFrequency_freeFrequencyPart_of_subset {n : }
      {J S : Finset (Fin n)} (hS : S  J) :
      FABL.liftFreeFrequency (FABL.freeFrequencyPart J S) = S
    theorem FABL.liftFreeFrequency_freeFrequencyPart_of_subset
      {n : } {J S : Finset (Fin n)}
      (hS : S  J) :
      FABL.liftFreeFrequency
          (FABL.freeFrequencyPart J S) =
        S
    Lift of the free part of `S` recovers `S` when `S ⊆ J`. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_sq_ambientRestrictionFourierCoeff {n : }
      (f : FABL.SignCube n  ) (J S : Finset (Fin n)) :
      (Finset.univ.expect fun z =>
          FABL.ambientRestrictionFourierCoeff f J S z ^ 2) =
        if S  J then
           T,
            FABL.fourierCoeff f
                (FABL.liftFreeFrequency (FABL.freeFrequencyPart J S) 
                  FABL.liftFixedFrequency T) ^
              2
        else 0
    theorem FABL.expect_sq_ambientRestrictionFourierCoeff
      {n : } (f : FABL.SignCube n  )
      (J S : Finset (Fin n)) :
      (Finset.univ.expect fun z =>
          FABL.ambientRestrictionFourierCoeff
              f J S z ^
            2) =
        if S  J then
           T,
            FABL.fourierCoeff f
                (FABL.liftFreeFrequency
                    (FABL.freeFrequencyPart J
                      S) 
                  FABL.liftFixedFrequency T) ^
              2
        else 0
    Conditional second moment via Corollary 3.22. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.subset_of_inter_eq {n : } {S U J : Finset (Fin n)}
      (h : U  J = S) : S  J
    theorem FABL.subset_of_inter_eq {n : }
      {S U J : Finset (Fin n)}
      (h : U  J = S) : S  J
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.subset_of_inter_eq_left {n : } {S U J : Finset (Fin n)}
      (h : U  J = S) : S  U
    theorem FABL.subset_of_inter_eq_left {n : }
      {S U J : Finset (Fin n)}
      (h : U  J = S) : S  U
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.sum_deltaRandomSubsetWeight_inter_eq (n : ) (δ : )
      (S U : Finset (Fin n)) :
      (∑ J, if U  J = S then FABL.deltaRandomSubsetWeight n δ J else 0) =
        if S  U then δ ^ S.card * (1 - δ) ^ (U \ S).card else 0
    theorem FABL.sum_deltaRandomSubsetWeight_inter_eq
      (n : ) (δ : ) (S U : Finset (Fin n)) :
      (∑ J,
          if U  J = S then
            FABL.deltaRandomSubsetWeight n δ J
          else 0) =
        if S  U then
          δ ^ S.card * (1 - δ) ^ (U \ S).card
        else 0
    Weight of free sets with fixed intersection `U ∩ J = S`. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.sum_sq_fourier_of_inter_eq {n : } (f : FABL.SignCube n  )
      (J S : Finset (Fin n)) (hS : S  J) :
       T,
          FABL.fourierCoeff f
              (FABL.liftFreeFrequency (FABL.freeFrequencyPart J S) 
                FABL.liftFixedFrequency T) ^
            2 =
         U, if U  J = S then FABL.fourierCoeff f U ^ 2 else 0
    theorem FABL.sum_sq_fourier_of_inter_eq {n : }
      (f : FABL.SignCube n  )
      (J S : Finset (Fin n)) (hS : S  J) :
       T,
          FABL.fourierCoeff f
              (FABL.liftFreeFrequency
                  (FABL.freeFrequencyPart J
                    S) 
                FABL.liftFixedFrequency T) ^
            2 =
         U,
          if U  J = S then
            FABL.fourierCoeff f U ^ 2
          else 0
    Ambient frequencies with free/fixed split relative to `J` matching free set `S`. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_sq_fourierCoeff_randomRestriction {n : }
      (f : FABL.SignCube n  ) (δ : ) (S : Finset (Fin n)) :
      (FABL.expectRandomRestriction n δ fun J z =>
          FABL.ambientRestrictionFourierCoeff f J S z ^ 2) =
         U,
          (if S  U then δ ^ S.card * (1 - δ) ^ (U \ S).card else 0) *
            FABL.fourierCoeff f U ^ 2
    theorem FABL.expect_sq_fourierCoeff_randomRestriction
      {n : } (f : FABL.SignCube n  )
      (δ : ) (S : Finset (Fin n)) :
      (FABL.expectRandomRestriction n δ
          fun J z =>
          FABL.ambientRestrictionFourierCoeff
              f J S z ^
            2) =
         U,
          (if S  U then
              δ ^ S.card *
                (1 - δ) ^ (U \ S).card
            else 0) *
            FABL.fourierCoeff f U ^ 2
    O'Donnell, Proposition 4.17 (second moment). 
Corollary4.3.4
Group: Chapter 4: DNF formulas and small-depth circuits (44)
Group member previews
Preview
Definition 4.1.1
Loading preview
Group member preview content is loaded from the rendered-fragment cache.
Statement uses 3
Statement dependency previews
Preview
Theorem 2.2.10
Loading preview
Statement dependency preview content is loaded from the rendered-fragment cache.
used by 0L∃∀N

Corollary 4.18. Fix f:\{-1,1\}^n\to\mathbb R and i\in[n]. If (\boldsymbol J\mid\boldsymbol z) is a \delta-random restriction, then \mathbb E[\operatorname{Inf}_i[f_{\boldsymbol J\mid\boldsymbol z}]] =\delta\operatorname{Inf}_i[f]. Hence also \mathbb E[\mathbf I[f_{\boldsymbol J\mid\boldsymbol z}]]=\delta\mathbf I[f]. A Fourier-free proof proceeds as in Exercise 4.9: condition on whether i is free. In the free case, the free/fixed product measure recovers the ambient influence, while \Pr[i\in\boldsymbol J]=\delta.

Lean code for Corollary4.3.48 declarations
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.sum_deltaRandomSubsetWeight_mem (n : ) (δ : ) (i : Fin n) :
      (∑ J, if i  J then FABL.deltaRandomSubsetWeight n δ J else 0) = δ
    theorem FABL.sum_deltaRandomSubsetWeight_mem
      (n : ) (δ : ) (i : Fin n) :
      (∑ J,
          if i  J then
            FABL.deltaRandomSubsetWeight n δ J
          else 0) =
        δ
    Weight of free sets containing a fixed coordinate equals `δ`. 
  • defdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    def FABL.freePart {n : } (J : Finset (Fin n)) (x : FABL.SignCube n) :
      FABL.FreeSignCube J
    def FABL.freePart {n : } (J : Finset (Fin n))
      (x : FABL.SignCube n) :
      FABL.FreeSignCube J
    Free assignment extracted from a full cube string. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.discreteDerivative_extendedSignRestriction_of_mem {n : }
      (f : FABL.SignCube n  ) (J : Finset (Fin n))
      (z : FABL.FixedSignCube J) (i : Fin n) (hi : i  J)
      (x : FABL.SignCube n) :
      (FABL.discreteDerivative i) (FABL.extendedSignRestriction f J z) x =
        (FABL.discreteDerivative i) f
          (FABL.combineSignCube J (fun j => x j) z)
    theorem FABL.discreteDerivative_extendedSignRestriction_of_mem
      {n : } (f : FABL.SignCube n  )
      (J : Finset (Fin n))
      (z : FABL.FixedSignCube J) (i : Fin n)
      (hi : i  J) (x : FABL.SignCube n) :
      (FABL.discreteDerivative i)
          (FABL.extendedSignRestriction f J z)
          x =
        (FABL.discreteDerivative i) f
          (FABL.combineSignCube J
            (fun j => x j) z)
    Free-coordinate discrete derivative of an extension equals that of `f`. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_sq_discreteDerivative_combine {n : }
      (f : FABL.SignCube n  ) (J : Finset (Fin n)) (i : Fin n)
      (z : FABL.FixedSignCube J) :
      (Finset.univ.expect fun x =>
          (FABL.discreteDerivative i) f
              (FABL.combineSignCube J (FABL.freePart J x) z) ^
            2) =
        Finset.univ.expect fun y =>
          (FABL.discreteDerivative i) f (FABL.combineSignCube J y z) ^ 2
    theorem FABL.expect_sq_discreteDerivative_combine
      {n : } (f : FABL.SignCube n  )
      (J : Finset (Fin n)) (i : Fin n)
      (z : FABL.FixedSignCube J) :
      (Finset.univ.expect fun x =>
          (FABL.discreteDerivative i) f
              (FABL.combineSignCube J
                (FABL.freePart J x) z) ^
            2) =
        Finset.univ.expect fun y =>
          (FABL.discreteDerivative i) f
              (FABL.combineSignCube J y z) ^
            2
    Averaging a squared discrete derivative that depends only on free coordinates. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_sq_discreteDerivative_product {n : }
      (f : FABL.SignCube n  ) (J : Finset (Fin n)) (i : Fin n) :
      (Finset.univ.expect fun y =>
          Finset.univ.expect fun z =>
            (FABL.discreteDerivative i) f (FABL.combineSignCube J y z) ^
              2) =
        FABL.influence f i
    theorem FABL.expect_sq_discreteDerivative_product
      {n : } (f : FABL.SignCube n  )
      (J : Finset (Fin n)) (i : Fin n) :
      (Finset.univ.expect fun y =>
          Finset.univ.expect fun z =>
            (FABL.discreteDerivative i) f
                (FABL.combineSignCube J y z) ^
              2) =
        FABL.influence f i
    Free/fixed product measure recovers ambient influence. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_influence_extendedSignRestriction_of_mem {n : }
      (f : FABL.SignCube n  ) (J : Finset (Fin n)) (i : Fin n)
      (hi : i  J) :
      (Finset.univ.expect fun z =>
          FABL.influence (FABL.extendedSignRestriction f J z) i) =
        FABL.influence f i
    theorem FABL.expect_influence_extendedSignRestriction_of_mem
      {n : } (f : FABL.SignCube n  )
      (J : Finset (Fin n)) (i : Fin n)
      (hi : i  J) :
      (Finset.univ.expect fun z =>
          FABL.influence
            (FABL.extendedSignRestriction f J
              z)
            i) =
        FABL.influence f i
    Conditional expected influence equals original when the coordinate is free. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_influence_extended_randomRestriction {n : }
      (f : FABL.SignCube n  ) (δ : ) (i : Fin n) :
      (FABL.expectRandomRestriction n δ fun J z =>
          FABL.influence (FABL.extendedSignRestriction f J z) i) =
        δ * FABL.influence f i
    theorem FABL.expect_influence_extended_randomRestriction
      {n : } (f : FABL.SignCube n  )
      (δ : ) (i : Fin n) :
      (FABL.expectRandomRestriction n δ
          fun J z =>
          FABL.influence
            (FABL.extendedSignRestriction f J
              z)
            i) =
        δ * FABL.influence f i
    O'Donnell, Corollary 4.18. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_totalInfluence_extended_randomRestriction {n : }
      (f : FABL.SignCube n  ) (δ : ) :
      (FABL.expectRandomRestriction n δ fun J z =>
          FABL.totalInfluence (FABL.extendedSignRestriction f J z)) =
        δ * FABL.totalInfluence f
    theorem FABL.expect_totalInfluence_extended_randomRestriction
      {n : } (f : FABL.SignCube n  )
      (δ : ) :
      (FABL.expectRandomRestriction n δ
          fun J z =>
          FABL.totalInfluence
            (FABL.extendedSignRestriction f J
              z)) =
        δ * FABL.totalInfluence f
    O'Donnell, Corollary 4.18 for total influence. 
Lemma4.3.5
Group: Chapter 4: DNF formulas and small-depth circuits (44)
Group member previews
Preview
Definition 4.1.1
Loading preview
Group member preview content is loaded from the rendered-fragment cache.
Statement uses 3
Statement dependency previews
Preview
Definition 4.1.1
Loading preview
Statement dependency preview content is loaded from the rendered-fragment cache.
used by 1L∃∀N

Lemma 4.19. Let T be a DNF term over \{-1,1\}^n and fix w\in\mathbb N^+. Let (\boldsymbol J\mid\boldsymbol z) be a (1/2)-random restriction on \{-1,1\}^n. Then \Pr\bigl[\operatorname{width}(T_{\boldsymbol J\mid\boldsymbol z})\ge w\bigr] \le\Bigl(\frac34\Bigr)^w. Under the half-random restriction, each coordinate is free with probability 1/2 and fixed to either sign with probability 1/4. Thus a term is not falsified with probability (3/4)^{\operatorname{width}(T)}, which gives the tail bound.

Lean code for Lemma4.3.54 declarations
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.literal_not_falsified_local_weight {n : } ( : FABL.Literal n) :
      (∑ c,
          FABL.coordRestrictionWeight c *
            if (.isFalsified fun x => c) = true then 0 else 1) =
        3 / 4
    theorem FABL.literal_not_falsified_local_weight
      {n : } ( : FABL.Literal n) :
      (∑ c,
          FABL.coordRestrictionWeight c *
            if
                (.isFalsified fun x => c) =
                  true then
              0
            else 1) =
        3 / 4
    Local weight that a literal is not fixed to False is `3/4`. 
  • defdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    def FABL.DNFTerm.notFalsified {n : } (T : FABL.DNFTerm n)
      (ρ : Fin n  FABL.CoordRestriction) : Bool
    def FABL.DNFTerm.notFalsified {n : }
      (T : FABL.DNFTerm n)
      (ρ : Fin n  FABL.CoordRestriction) :
      Bool
    Indicator that no literal of a term is falsified. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.term_not_falsified_weight {n : } (T : FABL.DNFTerm n) :
      (∑ ρ,
          FABL.restrictionAssignmentWeight ρ *
            if T.notFalsified ρ = true then 1 else 0) =
        (3 / 4) ^ T.width
    theorem FABL.term_not_falsified_weight {n : }
      (T : FABL.DNFTerm n) :
      (∑ ρ,
          FABL.restrictionAssignmentWeight ρ *
            if T.notFalsified ρ = true then 1
            else 0) =
        (3 / 4) ^ T.width
    Weight that a term is not falsified equals `(3/4)^{width}`. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.restrictedWidth_ge_probability_le {n : } (T : FABL.DNFTerm n)
      (w : ) :
      (∑ ρ,
          FABL.restrictionAssignmentWeight ρ *
            if w  T.restrictedWidthOf ρ then 1 else 0) 
        (3 / 4) ^ w
    theorem FABL.restrictedWidth_ge_probability_le
      {n : } (T : FABL.DNFTerm n) (w : ) :
      (∑ ρ,
          FABL.restrictionAssignmentWeight ρ *
            if w  T.restrictedWidthOf ρ then
              1
            else 0) 
        (3 / 4) ^ w
    O'Donnell, Lemma 4.19. 
Theorem4.3.6
Group: Chapter 4: DNF formulas and small-depth circuits (44)
Group member previews
Preview
Definition 4.1.1
Loading preview
Group member preview content is loaded from the rendered-fragment cache.
Statement uses 2
Statement dependency previews
Preview
Definition 4.1.4
Loading preview
Statement dependency preview content is loaded from the rendered-fragment cache.
used by 1L∃∀N

Theorem 4.20. Let f:\{-1,1\}^n\to\{-1,1\} be computable by a DNF of size s. Then \mathbf I[f]\le O(\log s). More precisely, \mathbf I[f]\le 2\bigl(\lceil\log_2 s\rceil+1\bigr), which implies the stated O(\log s) claim. Select one satisfied term on every true input, bound the number of negative pivotal coordinates by that term's width, prove the tail estimate \Pr[\operatorname{width}\ge k]\le s2^{-k}, and sum the tails.

Lean code for Theorem4.3.613 declarations
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.sum_inverse_two_pow_succ_Ico_le (m n : ) :
       k  Finset.Ico m n, (2 ^ (k + 1))⁻¹  (2 ^ m)⁻¹
    theorem FABL.sum_inverse_two_pow_succ_Ico_le
      (m n : ) :
       k  Finset.Ico m n, (2 ^ (k + 1))⁻¹ 
        (2 ^ m)⁻¹
  • defdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    def FABL.DNFFormula.selectedTerm {n : } (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n) : FABL.DNFTerm n
    def FABL.DNFFormula.selectedTerm {n : }
      (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n) : FABL.DNFTerm n
    A canonical satisfied term, with the empty term used when the DNF evaluates to `1`. 
  • defdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    def FABL.DNFFormula.selectedWidth {n : } (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n) : 
    def FABL.DNFFormula.selectedWidth {n : }
      (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n) : 
    Width of the selected satisfied term. 
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.DNFFormula.selectedTerm_mem {n : } (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n) (hx : φ.eval x = -1) :
      φ.selectedTerm x  φ.terms
    theorem FABL.DNFFormula.selectedTerm_mem {n : }
      (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n)
      (hx : φ.eval x = -1) :
      φ.selectedTerm x  φ.terms
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.DNFFormula.selectedTerm_eval {n : } (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n) (hx : φ.eval x = -1) :
      (φ.selectedTerm x).eval x = -1
    theorem FABL.DNFFormula.selectedTerm_eval {n : }
      (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n)
      (hx : φ.eval x = -1) :
      (φ.selectedTerm x).eval x = -1
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.DNFFormula.selectedWidth_eq_zero_of_eval_ne {n : }
      (φ : FABL.DNFFormula n) (x : FABL.SignCube n) (hx : φ.eval x  -1) :
      φ.selectedWidth x = 0
    theorem FABL.DNFFormula.selectedWidth_eq_zero_of_eval_ne
      {n : } (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n)
      (hx : φ.eval x  -1) :
      φ.selectedWidth x = 0
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.DNFFormula.selectedWidth_le_dimension {n : }
      (φ : FABL.DNFFormula n) (x : FABL.SignCube n) : φ.selectedWidth x  n
    theorem FABL.DNFFormula.selectedWidth_le_dimension
      {n : } (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n) :
      φ.selectedWidth x  n
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.card_negOnePivotal_le_selectedWidth {n : } (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n) :
      {i | FABL.IsNegOnePivotal φ.toBooleanFunction i x}.card 
        φ.selectedWidth x
    theorem FABL.card_negOnePivotal_le_selectedWidth
      {n : } (φ : FABL.DNFFormula n)
      (x : FABL.SignCube n) :
      {i |
            FABL.IsNegOnePivotal
              φ.toBooleanFunction i x}.card 
        φ.selectedWidth x
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.selectedWidth_tail_probability_le {n : } (φ : FABL.DNFFormula n)
      (k : ) (hk : 0 < k) :
      (FABL.uniformProbability fun x => k  φ.selectedWidth x) 
        φ.size * (2 ^ k)⁻¹
    theorem FABL.selectedWidth_tail_probability_le
      {n : } (φ : FABL.DNFFormula n) (k : )
      (hk : 0 < k) :
      (FABL.uniformProbability fun x =>
          k  φ.selectedWidth x) 
        φ.size * (2 ^ k)⁻¹
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.uniformProbability_le_one.{u_1} {Ω : Type u_1} [Fintype Ω]
      [Nonempty Ω] (P : Ω  Prop) [DecidablePred P] :
      FABL.uniformProbability P  1
    theorem FABL.uniformProbability_le_one.{u_1}
      {Ω : Type u_1} [Fintype Ω] [Nonempty Ω]
      (P : Ω  Prop) [DecidablePred P] :
      FABL.uniformProbability P  1
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_selectedWidth_eq_sum_tail {n : } (φ : FABL.DNFFormula n) :
      (Finset.univ.expect fun x => (φ.selectedWidth x)) =
         k  Finset.range n,
          FABL.uniformProbability fun x => k + 1  φ.selectedWidth x
    theorem FABL.expect_selectedWidth_eq_sum_tail
      {n : } (φ : FABL.DNFFormula n) :
      (Finset.univ.expect fun x =>
          (φ.selectedWidth x)) =
         k  Finset.range n,
          FABL.uniformProbability fun x =>
            k + 1  φ.selectedWidth x
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.expect_selectedWidth_le_clog_add_one {n : }
      (φ : FABL.DNFFormula n) {s : } (hsize : φ.size  s) :
      (Finset.univ.expect fun x => (φ.selectedWidth x)) 
        (Nat.clog 2 s) + 1
    theorem FABL.expect_selectedWidth_le_clog_add_one
      {n : } (φ : FABL.DNFFormula n) {s : }
      (hsize : φ.size  s) :
      (Finset.univ.expect fun x =>
          (φ.selectedWidth x)) 
        (Nat.clog 2 s) + 1
  • theoremdefined in FABL/Chapter04/RandomRestrictions.lean
    complete
    theorem FABL.totalInfluence_le_two_mul_clog_add_one_of_hasDNFSizeLE {n : }
      {f : FABL.BooleanFunction n} {s : } (hf : FABL.HasDNFSizeLE f s) :
      FABL.totalInfluence f.toReal  2 * ((Nat.clog 2 s) + 1)
    theorem FABL.totalInfluence_le_two_mul_clog_add_one_of_hasDNFSizeLE
      {n : } {f : FABL.BooleanFunction n}
      {s : } (hf : FABL.HasDNFSizeLE f s) :
      FABL.totalInfluence f.toReal 
        2 * ((Nat.clog 2 s) + 1)
    O'Donnell, Theorem 4.20, with an explicit logarithmic bound.