9.5. Applications of hypercontractivity
-
FABL.noiseOperator_one[complete] -
FABL.uniformLpNorm_two_noiseOperator_le_pow_of_degree[complete] -
FABL.lowDegree_uniformLpNorm_le[complete]
Theorem 9.21. If f:\{-1,1\}^n\to\mathbb R has degree at most k
and q\ge2, then
\|f\|_q\le(\sqrt{q-1})^k\|f\|_2.
Lean code for Theorem9.5.1●3 theorems
Associated Lean declarations
-
FABL.noiseOperator_one[complete]
-
FABL.uniformLpNorm_two_noiseOperator_le_pow_of_degree[complete]
-
FABL.lowDegree_uniformLpNorm_le[complete]
-
FABL.noiseOperator_one[complete] -
FABL.uniformLpNorm_two_noiseOperator_le_pow_of_degree[complete] -
FABL.lowDegree_uniformLpNorm_le[complete]
-
theoremdefined in FABL/Chapter09/ProductHypercontractivity.leancomplete
theorem FABL.noiseOperator_one {n : ℕ} (f : FABL.SignCube n → ℝ) : (FABL.noiseOperator 1) f = f
theorem FABL.noiseOperator_one {n : ℕ} (f : FABL.SignCube n → ℝ) : (FABL.noiseOperator 1) f = f
Noise at correlation one is the identity.
-
theoremdefined in FABL/Chapter09/ProductHypercontractivity.leancomplete
theorem FABL.uniformLpNorm_two_noiseOperator_le_pow_of_degree {n : ℕ} (rho : ℝ) (hrho : 1 ≤ rho) (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) : FABL.uniformLpNorm 2 ((FABL.noiseOperator rho) f) ≤ rho ^ k * FABL.uniformLpNorm 2 f
theorem FABL.uniformLpNorm_two_noiseOperator_le_pow_of_degree {n : ℕ} (rho : ℝ) (hrho : 1 ≤ rho) (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) : FABL.uniformLpNorm 2 ((FABL.noiseOperator rho) f) ≤ rho ^ k * FABL.uniformLpNorm 2 f
On a degree-`k` function, noise with parameter at least one grows the `L²` norm by at most the `k`th power of the parameter.
-
theoremdefined in FABL/Chapter09/ProductHypercontractivity.leancomplete
theorem FABL.lowDegree_uniformLpNorm_le {n : ℕ} (q : ℝ) (hq : 2 ≤ q) (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) : FABL.uniformLpNorm q f ≤ √(q - 1) ^ k * FABL.uniformLpNorm 2 f
theorem FABL.lowDegree_uniformLpNorm_le {n : ℕ} (q : ℝ) (hq : 2 ≤ q) (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) : FABL.uniformLpNorm q f ≤ √(q - 1) ^ k * FABL.uniformLpNorm 2 f
O'Donnell, Theorem 9.21: low-degree `L^q` norm growth.
-
FABL.lowDegree_two_le_expFactor_mul[complete] -
FABL.lowDegree_two_le_exp_pow_mul_one[complete]
Theorem 9.22. If \deg(f)\le k, then
\|f\|_2\le e^k\|f\|_1.
More generally, for 1\le p\le2,
\|f\|_2
\le\left(e^{2/p-1}\right)^k\|f\|_p.
Lean code for Theorem9.5.2●2 theorems
Associated Lean declarations
-
FABL.lowDegree_two_le_expFactor_mul[complete]
-
FABL.lowDegree_two_le_exp_pow_mul_one[complete]
-
FABL.lowDegree_two_le_expFactor_mul[complete] -
FABL.lowDegree_two_le_exp_pow_mul_one[complete]
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.lowDegree_two_le_expFactor_mul {n : ℕ} (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) (p : ℝ) (hp1 : 1 ≤ p) (hp2 : p ≤ 2) : FABL.uniformLpNorm 2 f ≤ (Real.exp 1 ^ (2 / p - 1)) ^ k * FABL.uniformLpNorm p f
theorem FABL.lowDegree_two_le_expFactor_mul {n : ℕ} (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) (p : ℝ) (hp1 : 1 ≤ p) (hp2 : p ≤ 2) : FABL.uniformLpNorm 2 f ≤ (Real.exp 1 ^ (2 / p - 1)) ^ k * FABL.uniformLpNorm p f
O'Donnell, Theorem 9.22, with the complete sharp-in-this-proof factor for every `1 ≤ p ≤ 2`, including the `p = 1` endpoint.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.lowDegree_two_le_exp_pow_mul_one {n : ℕ} (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) : FABL.uniformLpNorm 2 f ≤ Real.exp 1 ^ k * FABL.uniformLpNorm 1 f
theorem FABL.lowDegree_two_le_exp_pow_mul_one {n : ℕ} (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) : FABL.uniformLpNorm 2 f ≤ Real.exp 1 ^ k * FABL.uniformLpNorm 1 f
The `p = 1` endpoint of Theorem 9.22.
-
FABL.fourierDegree_linearForm_le_one[complete] -
FABL.khintchine_of_le_two[complete] -
FABL.khintchine_of_two_le[complete] -
FABL.khintchineInequality[complete]
Khintchine's Inequality. For
L(x)=\sum_i a_ix_i and every p\in[1,\infty), there are constants
0<c_p<C_p<\infty, depending only on p, such that
c_p\|L\|_2\le\|L\|_p\le C_p\|L\|_2.
Lean code for Theorem9.5.3●4 theorems
Associated Lean declarations
-
FABL.fourierDegree_linearForm_le_one[complete]
-
FABL.khintchine_of_le_two[complete]
-
FABL.khintchine_of_two_le[complete]
-
FABL.khintchineInequality[complete]
-
FABL.fourierDegree_linearForm_le_one[complete] -
FABL.khintchine_of_le_two[complete] -
FABL.khintchine_of_two_le[complete] -
FABL.khintchineInequality[complete]
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.fourierDegree_linearForm_le_one {n : ℕ} (a : Fin n → ℝ) : FABL.fourierDegree (FABL.linearForm a) ≤ 1
theorem FABL.fourierDegree_linearForm_le_one {n : ℕ} (a : Fin n → ℝ) : FABL.fourierDegree (FABL.linearForm a) ≤ 1
A Rademacher linear form has Fourier degree at most one.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.khintchine_of_le_two {n : ℕ} (a : Fin n → ℝ) (p : ℝ) (hp1 : 1 ≤ p) (hp2 : p ≤ 2) : Real.exp (-(2 / p - 1)) * FABL.uniformLpNorm 2 (FABL.linearForm a) ≤ FABL.uniformLpNorm p (FABL.linearForm a) ∧ FABL.uniformLpNorm p (FABL.linearForm a) ≤ FABL.uniformLpNorm 2 (FABL.linearForm a)
theorem FABL.khintchine_of_le_two {n : ℕ} (a : Fin n → ℝ) (p : ℝ) (hp1 : 1 ≤ p) (hp2 : p ≤ 2) : Real.exp (-(2 / p - 1)) * FABL.uniformLpNorm 2 (FABL.linearForm a) ≤ FABL.uniformLpNorm p (FABL.linearForm a) ∧ FABL.uniformLpNorm p (FABL.linearForm a) ≤ FABL.uniformLpNorm 2 (FABL.linearForm a)
Khintchine's inequality for `1 ≤ p ≤ 2`, with explicit constants.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.khintchine_of_two_le {n : ℕ} (a : Fin n → ℝ) (p : ℝ) (hp : 2 ≤ p) : FABL.uniformLpNorm 2 (FABL.linearForm a) ≤ FABL.uniformLpNorm p (FABL.linearForm a) ∧ FABL.uniformLpNorm p (FABL.linearForm a) ≤ √(p - 1) * FABL.uniformLpNorm 2 (FABL.linearForm a)
theorem FABL.khintchine_of_two_le {n : ℕ} (a : Fin n → ℝ) (p : ℝ) (hp : 2 ≤ p) : FABL.uniformLpNorm 2 (FABL.linearForm a) ≤ FABL.uniformLpNorm p (FABL.linearForm a) ∧ FABL.uniformLpNorm p (FABL.linearForm a) ≤ √(p - 1) * FABL.uniformLpNorm 2 (FABL.linearForm a)
Khintchine's inequality for `p ≥ 2`, with explicit constants.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.khintchineInequality {n : ℕ} (a : Fin n → ℝ) (p : ℝ) (hp : 1 ≤ p) : ∃ c C, 0 < c ∧ c < C ∧ c * FABL.uniformLpNorm 2 (FABL.linearForm a) ≤ FABL.uniformLpNorm p (FABL.linearForm a) ∧ FABL.uniformLpNorm p (FABL.linearForm a) ≤ C * FABL.uniformLpNorm 2 (FABL.linearForm a)
theorem FABL.khintchineInequality {n : ℕ} (a : Fin n → ℝ) (p : ℝ) (hp : 1 ≤ p) : ∃ c C, 0 < c ∧ c < C ∧ c * FABL.uniformLpNorm 2 (FABL.linearForm a) ≤ FABL.uniformLpNorm p (FABL.linearForm a) ∧ FABL.uniformLpNorm p (FABL.linearForm a) ≤ C * FABL.uniformLpNorm 2 (FABL.linearForm a)
O'Donnell's existential formulation of Khintchine's Inequality.
-
FABL.lowDegree_tail_bound_parameter[complete] -
FABL.lowDegree_tail_bound[complete]
Theorem 9.23. If \deg(f)\le k, \|f\|_2>0, and
t\ge(2e)^{k/2}, then
\Pr[|f(x)|\ge t\|f\|_2]
\le\exp\left(-\frac{k}{2e}t^{2/k}\right).
The positive-norm condition is necessary for the printed non-strict event:
for the zero function, |f|\ge t\|f\|_2 is the whole cube while the
displayed upper bound is strictly below one.
Lean code for Theorem9.5.4●2 theorems
Associated Lean declarations
-
FABL.lowDegree_tail_bound_parameter[complete]
-
FABL.lowDegree_tail_bound[complete]
-
FABL.lowDegree_tail_bound_parameter[complete] -
FABL.lowDegree_tail_bound[complete]
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.lowDegree_tail_bound_parameter {n : ℕ} (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) (q t : ℝ) (hq : 2 ≤ q) (ht : 0 < t) (hnorm : 0 < FABL.uniformLpNorm 2 f) : (FABL.uniformProbability fun x => t * FABL.uniformLpNorm 2 f ≤ |f x|) ≤ (q ^ (↑k / 2) / t) ^ q
theorem FABL.lowDegree_tail_bound_parameter {n : ℕ} (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) (q t : ℝ) (hq : 2 ≤ q) (ht : 0 < t) (hnorm : 0 < FABL.uniformLpNorm 2 f) : (FABL.uniformProbability fun x => t * FABL.uniformLpNorm 2 f ≤ |f x|) ≤ (q ^ (↑k / 2) / t) ^ q
Markov plus Theorem 9.21, with a free moment parameter.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.lowDegree_tail_bound {n : ℕ} (f : FABL.SignCube n → ℝ) (k : ℕ) (hk : 0 < k) (hdegree : FABL.fourierDegree f ≤ k) (t : ℝ) (ht : (2 * Real.exp 1) ^ (↑k / 2) ≤ t) (hnorm : 0 < FABL.uniformLpNorm 2 f) : (FABL.uniformProbability fun x => t * FABL.uniformLpNorm 2 f ≤ |f x|) ≤ Real.exp (-↑k / (2 * Real.exp 1) * t ^ (2 / ↑k))
theorem FABL.lowDegree_tail_bound {n : ℕ} (f : FABL.SignCube n → ℝ) (k : ℕ) (hk : 0 < k) (hdegree : FABL.fourierDegree f ≤ k) (t : ℝ) (ht : (2 * Real.exp 1) ^ (↑k / 2) ≤ t) (hnorm : 0 < FABL.uniformLpNorm 2 f) : (FABL.uniformProbability fun x => t * FABL.uniformLpNorm 2 f ≤ |f x|) ≤ Real.exp (-↑k / (2 * Real.exp 1) * t ^ (2 / ↑k))
O'Donnell, Theorem 9.23, with the necessary nonzero-function condition made explicit for the book's non-strict tail event.
Theorem 9.24. If f is nonconstant and \deg(f)\le k, then
\Pr[f(x)>\mathbb E f]
\ge\frac14e^{-2k}.
Lean code for Theorem9.5.5●1 theorem
Associated Lean declarations
-
FABL.lowDegree_aboveMean_probability[complete]
-
FABL.lowDegree_aboveMean_probability[complete]
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.lowDegree_aboveMean_probability {n : ℕ} (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) (hnonconstant : ¬∃ c, ∀ (x : FABL.SignCube n), f x = c) : 1 / 4 * Real.exp (-2 * ↑k) ≤ FABL.uniformProbability fun x => FABL.mean f < f x
theorem FABL.lowDegree_aboveMean_probability {n : ℕ} (f : FABL.SignCube n → ℝ) (k : ℕ) (hdegree : FABL.fourierDegree f ≤ k) (hnonconstant : ¬∃ c, ∀ (x : FABL.SignCube n), f x = c) : 1 / 4 * Real.exp (-2 * ↑k) ≤ FABL.uniformProbability fun x => FABL.mean f < f x
O'Donnell, Theorem 9.24: a nonconstant low-degree function exceeds its mean with one-sided probability at least `(1/4) exp(-2k)`.
-
FABL.uniformLpNorm_zeroOne[complete] -
FABL.noiseStability_le_uniformLpNorm_oneAdd_sq[complete] -
FABL.smallSetExpansion[complete] -
FABL.smallSetExpansion_event[complete] -
FABL.smallSetExpansion_conditional[complete]
Small-Set Expansion Theorem. If A\subseteq\{-1,1\}^n has volume
\alpha, then for 0\le\rho\le1,
\operatorname{Stab}_\rho[1_A]
\le\alpha^{2/(1+\rho)}.
For \alpha>0, equivalently,
\Pr_{x\sim A,\,y\sim N_\rho(x)}[y\in A]
\le\alpha^{(1-\rho)/(1+\rho)}.
Lean code for Theorem9.5.6●5 theorems
Associated Lean declarations
-
FABL.uniformLpNorm_zeroOne[complete]
-
FABL.noiseStability_le_uniformLpNorm_oneAdd_sq[complete]
-
FABL.smallSetExpansion[complete]
-
FABL.smallSetExpansion_event[complete]
-
FABL.smallSetExpansion_conditional[complete]
-
FABL.uniformLpNorm_zeroOne[complete] -
FABL.noiseStability_le_uniformLpNorm_oneAdd_sq[complete] -
FABL.smallSetExpansion[complete] -
FABL.smallSetExpansion_event[complete] -
FABL.smallSetExpansion_conditional[complete]
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.uniformLpNorm_zeroOne {n : ℕ} (f : FABL.SignCube n → ℝ) (α p : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hp : 0 < p) : FABL.uniformLpNorm p f = α ^ p⁻¹
theorem FABL.uniformLpNorm_zeroOne {n : ℕ} (f : FABL.SignCube n → ℝ) (α p : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hp : 0 < p) : FABL.uniformLpNorm p f = α ^ p⁻¹
Exact `L^p` norm of a zero-one function.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.noiseStability_le_uniformLpNorm_oneAdd_sq {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc 0 1) (f : FABL.SignCube n → ℝ) : FABL.noiseStability ρ ⋯ f ≤ FABL.uniformLpNorm (1 + ρ) f ^ 2
theorem FABL.noiseStability_le_uniformLpNorm_oneAdd_sq {n : ℕ} (ρ : ℝ) (hρ : ρ ∈ Set.Icc 0 1) (f : FABL.SignCube n → ℝ) : FABL.noiseStability ρ ⋯ f ≤ FABL.uniformLpNorm (1 + ρ) f ^ 2
General `(p,2)` stability form on the Boolean cube.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.smallSetExpansion {n : ℕ} (f : FABL.SignCube n → ℝ) (α ρ : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hρ : ρ ∈ Set.Icc 0 1) : FABL.noiseStability ρ ⋯ f ≤ α ^ (2 / (1 + ρ))
theorem FABL.smallSetExpansion {n : ℕ} (f : FABL.SignCube n → ℝ) (α ρ : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hρ : ρ ∈ Set.Icc 0 1) : FABL.noiseStability ρ ⋯ f ≤ α ^ (2 / (1 + ρ))
O'Donnell's Small-Set Expansion Theorem, in noise-stability form.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.smallSetExpansion_event {n : ℕ} (f : FABL.SignCube n → ℝ) (α ρ : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hρ : ρ ∈ Set.Icc 0 1) : (FABL.pmfEventProbability (FABL.correlatedPairPMF ρ ⋯) fun xy => f xy.1 = 1 ∧ f xy.2 = 1) ≤ α ^ (2 / (1 + ρ))
theorem FABL.smallSetExpansion_event {n : ℕ} (f : FABL.SignCube n → ℝ) (α ρ : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hρ : ρ ∈ Set.Icc 0 1) : (FABL.pmfEventProbability (FABL.correlatedPairPMF ρ ⋯) fun xy => f xy.1 = 1 ∧ f xy.2 = 1) ≤ α ^ (2 / (1 + ρ))
Small-Set Expansion in joint-event form.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.smallSetExpansion_conditional {n : ℕ} (f : FABL.SignCube n → ℝ) (α ρ : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hα : 0 < α) (hρ : ρ ∈ Set.Icc 0 1) : (FABL.pmfEventProbability (FABL.correlatedPairPMF ρ ⋯) fun xy => f xy.1 = 1 ∧ f xy.2 = 1) / α ≤ α ^ ((1 - ρ) / (1 + ρ))
theorem FABL.smallSetExpansion_conditional {n : ℕ} (f : FABL.SignCube n → ℝ) (α ρ : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hα : 0 < α) (hρ : ρ ∈ Set.Icc 0 1) : (FABL.pmfEventProbability (FABL.correlatedPairPMF ρ ⋯) fun xy => f xy.1 = 1 ∧ f xy.2 = 1) / α ≤ α ^ ((1 - ρ) / (1 + ρ))
Equivalent conditional-probability form of the Small-Set Expansion Theorem.
Corollary 9.25. For every Boolean-valued f, coordinate i, and
0\le\rho\le1,
\operatorname{Inf}_i^{(\rho)}[f]
\le\operatorname{Inf}_i[f]^{2/(1+\rho)}.
Lean code for Theorem9.5.7●1 theorem
Associated Lean declarations
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.stableInfluence_le_booleanInfluence_rpow {n : ℕ} (f : FABL.BooleanFunction n) (i : Fin n) (ρ : ℝ) (hρ : ρ ∈ Set.Icc 0 1) : FABL.stableInfluence ρ f.toReal i ≤ FABL.booleanInfluence f i ^ (2 / (1 + ρ))
theorem FABL.stableInfluence_le_booleanInfluence_rpow {n : ℕ} (f : FABL.BooleanFunction n) (i : Fin n) (ρ : ℝ) (hρ : ρ ∈ Set.Icc 0 1) : FABL.stableInfluence ρ f.toReal i ≤ FABL.booleanInfluence f i ^ (2 / (1 + ρ))
O'Donnell, Corollary 9.25: general stable-influence bound.
-
FABL.rho_pow_mul_fourierWeightAtMost_le_noiseStability[complete] -
FABL.fourierWeightAtMost_le_invPow_mul_rpow[complete] -
FABL.levelKInequality[complete] -
FABL.levelOneInequality_sharp[complete]
Level-k Inequalities. Let
f:\{-1,1\}^n\to\{0,1\} have mean \alpha, and let
k\in\mathbb N^+ satisfy k\le2\ln(1/\alpha). Then
W^{\le k}[f]
\le
\left(\frac{2e}{k}\ln\frac1\alpha\right)^k\alpha^2.
At level one the sharp form is
W^1[f]\le2\alpha^2\ln(1/\alpha).
Lean code for Theorem9.5.8●4 theorems
Associated Lean declarations
-
FABL.rho_pow_mul_fourierWeightAtMost_le_noiseStability[complete]
-
FABL.fourierWeightAtMost_le_invPow_mul_rpow[complete]
-
FABL.levelKInequality[complete]
-
FABL.levelOneInequality_sharp[complete]
-
FABL.rho_pow_mul_fourierWeightAtMost_le_noiseStability[complete] -
FABL.fourierWeightAtMost_le_invPow_mul_rpow[complete] -
FABL.levelKInequality[complete] -
FABL.levelOneInequality_sharp[complete]
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.rho_pow_mul_fourierWeightAtMost_le_noiseStability {n : ℕ} (k : ℕ) (f : FABL.SignCube n → ℝ) (ρ : ℝ) (hρ : ρ ∈ Set.Ioc 0 1) : ρ ^ k * FABL.fourierWeightAtMost k f ≤ FABL.noiseStability ρ ⋯ f
theorem FABL.rho_pow_mul_fourierWeightAtMost_le_noiseStability {n : ℕ} (k : ℕ) (f : FABL.SignCube n → ℝ) (ρ : ℝ) (hρ : ρ ∈ Set.Ioc 0 1) : ρ ^ k * FABL.fourierWeightAtMost k f ≤ FABL.noiseStability ρ ⋯ f
Low-degree Fourier mass is controlled by noise stability at every positive correlation.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.fourierWeightAtMost_le_invPow_mul_rpow {n : ℕ} (k : ℕ) (f : FABL.SignCube n → ℝ) (α ρ : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hρ : ρ ∈ Set.Ioc 0 1) : FABL.fourierWeightAtMost k f ≤ ρ⁻¹ ^ k * α ^ (2 / (1 + ρ))
theorem FABL.fourierWeightAtMost_le_invPow_mul_rpow {n : ℕ} (k : ℕ) (f : FABL.SignCube n → ℝ) (α ρ : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (hρ : ρ ∈ Set.Ioc 0 1) : FABL.fourierWeightAtMost k f ≤ ρ⁻¹ ^ k * α ^ (2 / (1 + ρ))
Direct small-set bound on low-degree mass at an arbitrary positive correlation.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.levelKInequality {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (k : ℕ) (hkpos : 0 < k) (hk : ↑k ≤ 2 * Real.log (1 / α)) : FABL.fourierWeightAtMost k f ≤ (2 * Real.exp 1 / ↑k * Real.log (1 / α)) ^ k * α ^ 2
theorem FABL.levelKInequality {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) (k : ℕ) (hkpos : 0 < k) (hk : ↑k ≤ 2 * Real.log (1 / α)) : FABL.fourierWeightAtMost k f ≤ (2 * Real.exp 1 / ↑k * Real.log (1 / α)) ^ k * α ^ 2
O'Donnell's Level-`k` Inequalities with the explicit book constant.
-
theoremdefined in FABL/Chapter09/HypercontractiveApplications.leancomplete
theorem FABL.levelOneInequality_sharp {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) : FABL.fourierWeightAtLevel 1 f ≤ 2 * α ^ 2 * Real.log (1 / α)
theorem FABL.levelOneInequality_sharp {n : ℕ} (f : FABL.SignCube n → ℝ) (α : ℝ) (hvalues : ∀ (x : FABL.SignCube n), f x = 0 ∨ f x = 1) (hmean : FABL.mean f = α) : FABL.fourierWeightAtLevel 1 f ≤ 2 * α ^ 2 * Real.log (1 / α)
The sharp level-one inequality in the book, for every zero-one density.