On this page
Gaussian concentration for Lipschitz functions
Imports 10 library modules and 29 external
- GaussianAnalysis.gaussian.measurable_matrix
- GaussianAnalysis.gaussian.normal_integral
- GaussianAnalysis.gaussian.gauss
- GaussianAnalysis.gaussian.gauss_ibp
- GaussianAnalysis.gaussian.gauss_interpolation
- GaussianAnalysis.matrix.euclidean
- GaussianAnalysis.gaussian.partial_derivative
- GaussianAnalysis.comparison.normal_mgf
- GaussianAnalysis.comparison.gauss_rotation
- GaussianAnalysis.comparison.gauss_smoothing
External: HB.structures, mathcomp.classical.all_ssreflect_compat, mathcomp.algebra.algebraic_hierarchy.ssralg, mathcomp.algebra.numeric_hierarchy.ssrnum, mathcomp.algebra.interval, mathcomp.algebra.matrix, mathcomp.algebra.ring_tactic, mathcomp.algebra.field_tactic, mathcomp.algebra.arithmetic_tactic, mathcomp.classical.boolp, mathcomp.classical.classical_sets, mathcomp.classical.functions, mathcomp.reals.reals, mathcomp.analysis.ereal, mathcomp.analysis.topology_theory.topology, mathcomp.analysis.normedtype_theory.normedtype, mathcomp.analysis.sequences, mathcomp.analysis.derive, mathcomp.analysis.realfun, mathcomp.analysis.exp, mathcomp.analysis.trigo, mathcomp.analysis.measure_theory.measure, mathcomp.analysis.lebesgue_stieltjes_measure, mathcomp.analysis.lebesgue_measure, mathcomp.analysis.numfun, mathcomp.analysis.measurable_realfun, mathcomp.analysis.lebesgue_integral_theory.lebesgue_integral, mathcomp.analysis.probability_theory.probability, mathcomp.analysis.probability_theory.normal_distribution
Imported by 10
- GaussianAnalysis.matrix.covering
- GaussianAnalysis.matrix.tensor_packing
- GaussianAnalysis.random_matrix.goe
- GaussianAnalysis.random_matrix.goe_edge
- GaussianAnalysis.random_matrix.wigner
- GaussianAnalysis.random_matrix.matrix_series
- GaussianAnalysis.comparison.escape
- GaussianAnalysis.comparison.cgmt
- GaussianAnalysis.comparison.cone_escape
- GaussianAnalysis.comparison.sudakov
Required by 4 files of the paper proofs
Let x have law gauss n and let F : 'rV[R]_n -> R be L-Lipschitz for the Euclidean norm: |F x - F y| <= L |x - y|. Then E[expR (t (F x - E[F]))] <= expR (t^2 L^2 / 2) for every real t (gauss_concentration), with the sharp constant of Tsirelson, Ibragimov and Sudakov. For u >= 0, P(F x >= E[F] + u) and P(F x <= E[F] - u) are at most expR (- u^2 / (2 L^2)) (gauss_concentration_upper_tail, gauss_concentration_lower_tail), and P(|F x - E[F]| >= u) is at most twice that (gauss_concentration_abs_tail). The variance of F x is at most L^2: this is the Gaussian Poincare inequality for Lipschitz functions (gauss_poincare_lipschitz), from the bound on E[expR (t (F x - E[F]))] at t and - t, with cosh y >= 1 + y^2 / 2 (expR_cosh_ge), as t -> 0.
The proof follows Herbst. Let F be bounded, of class C^1, with gradient of Euclidean norm at most L, and let t >= 0. Let y be an independent copy of x and phi th = E[F x expR (t F (cos th x + sin th y))]. After the rotation of (x, y) by th, Gaussian integration by parts in y (gauss2_ibp_snd) shows that the derivative of phi is at least - sin th t L^2 E[expR (t F)]. Comparing phi 0 and phi (pi / 2) gives the covariance inequality E[F expR (t F)] <= (E[F] + t L^2) E[expR (t F)] (gauss_herbst_ineq), and integrating the resulting differential inequality for t |-> E[expR (t F)] gives the bound for such F (gauss_concentration_smooth). A general Lipschitz F is the pointwise limit of Gaussian smoothings (gauss_smoothing.v) of F clipped to [-(k + 1), k + 1] (clip), and Fatou's lemma concludes.
The file also proves facts on the law gauss n \x gauss n of two independent standard Gaussian vectors: its marginals (integral_gauss2_fst, integral_gauss2_snd), the independence of its components (integral_gauss2_prod) and integration by parts in the second component (gauss2_ibp_snd).
clip k x | the point of [-k, k] closest to x |
Main results
| Results | Statement | Used by theories/ |
|---|---|---|
For x of law gauss n, F bounded of class C^1 with gradient norm at most L, and t >= 0: E[F expR (t F)] <= (E[F] + t L^2) E[expR (t F)]. | ||
Gaussian concentration (Tsirelson, Ibragimov, Sudakov): if F is L-Lipschitz for the Euclidean norm, then E[expR (t (F x - E F))] <= expR (t^2 L^2 / 2) for every real t. | Gaussian/ | |
For such F and u >= 0, P(F x >= E F + u) and P(F x <= E F - u) are at most expR (- u^2 / (2 L^2)), and P(|F x - E F| >= u) is at most twice that. | ||
Gaussian Poincaré inequality for Lipschitz functions: if F is L-Lipschitz for the Euclidean norm and x has law gauss n, then E[(F x - E F)^2] <= L^2. | Gaussian/ | |
2 + x^2 <= expR x + expR (- x) for every real x, and 2 y <= expR y - expR (- y) for y >= 0. | ||
For (x, y) of law gauss n \x gauss n: the marginals are gauss n, the components are independent, and E[y_j H(x, y)] = E[d_(y_j) H(x, y)]. |
Declarations26 exported, 50 local
Context d (T : measurableType d) (R : realType) (mu : probability T R).
forall [d : measure_display] [T : measurableType d] [R : realType] (mu : probability T R) [f : T -> R] [C : R], measurable_fun [set: T] f -> (forall x : T, `|f x| <= C) -> mu.-integrable [set: T] (EFin \o f)
Statement as written in the source
Lemma integrable_prob_bounded (f : T -> R) (C : R) : measurable_fun [set: T] f -> (forall x, `|f x| <= C) -> mu.-integrable [set: T] (EFin \o f).
Used by 11 declarations in 2 files
- this file: integral_gauss2_prod, gauss2_ibp_snd, herbst_phi_derive, herbst_df_lower, gauss_herbst_ineq, herbst_mgf_derive, gauss_concentration_smooth_ge0, gauss_concentration_smooth, gauss_concentration
- GaussianAnalysis.comparison.sudakov: gauss_max_lower, dual_sudakov
Context {R : realType} {n : nat}. Local Notation P := (gauss n \x gauss n)%E.
forall {R : realType} {n : nat} [g : 'rV_n -> R], (gauss n).-integrable [set: 'rV_n] (EFin \o g) -> (gauss n \x gauss n)%E.-integrable [set: 'rV_n * 'rV_n] (fun p : ('rV_n * 'rV_n) => (g p.1)%:E)
Statement as written in the source
Lemma integrable_gauss2_fst (g : 'rV[R]_n -> R) : (gauss n).-integrable [set: 'rV[R]_n] (EFin \o g) -> P.-integrable [set: 'rV[R]_n * 'rV[R]_n] (fun p => (g p.1)%:E).
Used by 3 declarations in 1 file
- this file: integral_gauss2_fst, integrable_herbst_D, herbst_df_lower
forall {R : realType} {n : nat} [g : 'rV_n -> R], (gauss n).-integrable [set: 'rV_n] (EFin \o g) -> (gauss n \x gauss n)%E.-integrable [set: 'rV_n * 'rV_n] (fun p : ('rV_n * 'rV_n) => (g p.2)%:E)
Statement as written in the source
Lemma integrable_gauss2_snd (g : 'rV[R]_n -> R) : (gauss n).-integrable [set: 'rV[R]_n] (EFin \o g) -> P.-integrable [set: 'rV[R]_n * 'rV[R]_n] (fun p => (g p.2)%:E).
Used by 4 declarations in 1 file
- this file: integral_gauss2_snd, gauss2_ibp_snd, integrable_herbst_D, herbst_df_lower
forall {R : realType} {n : nat} [g : 'rV_n -> R], (gauss n).-integrable [set: 'rV_n] (EFin \o g) -> (\int[gauss n \x gauss n]_p (g p.1)%:E)%E = (\int[gauss n]_x (g x)%:E)%E
For (x, y) of law gauss n \x gauss n: the marginals are gauss n, the components are independent, and E[y_j H(x, y)] = E[d_(y_j) H(x, y)].
Statement as written in the source
Used by 2 declarations in 1 file
- this file: herbst_df_lower, gauss_herbst_ineq
forall {R : realType} {n : nat} [f g : 'rV_n -> R] [C : R], measurable_fun [set: 'rV_n] f -> measurable_fun [set: 'rV_n] g -> (forall x : 'rV_n, `|f x| <= C) -> (forall x : 'rV_n, `|g x| <= C) -> (\int[gauss n \x gauss n]_p (f p.1 * g p.2)%:E)%E = (\int[gauss n]_x (f x)%:E * \int[gauss n]_y (g y)%:E)%E
The two components of a vector of law gauss n \x gauss n are independent.
Stated in the README together with integral_gauss2_fst.
Statement as written in the source
Used by 1 declaration in 1 file
- this file: gauss_herbst_ineq
forall {R : realType} {n : nat} [j : 'I_n] [H dH : 'rV_n * 'rV_n -> R] [C : R], measurable_fun [set: 'rV_n * 'rV_n] H -> measurable_fun [set: 'rV_n * 'rV_n] dH -> (forall (x : 'rV_n) (v : 'rV[R]_n), is_derive v (delta_mx 0 j) (fun v0 : 'rV[R]_n => H (x, v0)) (dH (x, v))) -> (forall x : 'rV_n, continuous (fun v : 'rV[R]_n => H (x, v))) -> (forall x : 'rV_n, continuous (fun v : 'rV[R]_n => dH (x, v))) -> (forall p : 'rV_n * 'rV_n, `|H p| <= C) -> (forall p : 'rV_n * 'rV_n, `|dH p| <= C) -> (\int[gauss n \x gauss n]_p (p.2 0 j * H p)%:E)%E = (\int[gauss n \x gauss n]_p (dH p)%:E)%E
Gaussian integration by parts in the second component.
Stated in the README together with integral_gauss2_fst.
Statement as written in the source
Lemma gauss2_ibp_snd (j : 'I_n) (H dH : 'rV[R]_n * 'rV[R]_n -> R) (C : R) : measurable_fun [set: 'rV[R]_n * 'rV[R]_n] H -> measurable_fun [set: 'rV[R]_n * 'rV[R]_n] dH -> (forall x v, is_derive v (delta_mx 0 j) (fun v => H (x, v)) (dH (x, v))) -> (forall x, continuous (fun v => H (x, v))) -> (forall x, continuous (fun v => dH (x, v))) -> (forall p, `|H p| <= C) -> (forall p, `|dH p| <= C) -> (\int[P]_p (p.2 0 j * H p)%:E = \int[P]_p (dH p)%:E)%E.
Used by 1 declaration in 1 file
- this file: herbst_ibp
F is bounded by K, of class C^1 with partial derivatives DF j, and its gradient has Euclidean norm at most L.
Context {R : realType} {n : nat}. Variables (F : 'rV[R]_n -> R) (DF : 'I_n -> 'rV[R]_n -> R) (K L : R). Hypothesis dF : forall j x, is_derive x (delta_mx 0 j) F (DF j x). Hypothesis cDF : forall j, continuous (DF j). Hypothesis FK : forall x, `|F x| <= K. Hypothesis gradL : forall x, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2. Local Notation P := (gauss n \x gauss n)%E.
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> forall x : 'rV[R]_n, differentiable (F : 'rV_n -> R^o) x
Statement as written in the source
Local Lemma herbst_Fdiff x : differentiable (F : 'rV[R]_n -> R^o) x.
Used by 3 declarations in 1 file
- this file: herbst_cF, herbst_is_derive_Fz, herbst_is_derive_Fr
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> continuous F
Statement as written in the source
Local Lemma herbst_cF : continuous F.
Used by 2 declarations in 1 file
- this file: herbst_mF, herbst_ibp
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> measurable_fun [set: 'rV_n] F
Statement as written in the source
Local Lemma herbst_mF : measurable_fun [set: 'rV[R]_n] F.
Used by 8 declarations in 1 file
forall {R : realType} {n : nat} [DF : 'I_n -> 'rV_n -> R], (forall j : 'I_n, continuous (DF j)) -> forall j : 'I_n, measurable_fun [set: 'rV_n] (DF j)
Statement as written in the source
Local Lemma herbst_mDF j : measurable_fun [set: 'rV[R]_n] (DF j).
Used by 4 declarations in 1 file
- this file: herbst_mdf, herbst_mk, herbst_ibp, herbst_df_lower
forall {R : realType} {n : nat} [F : 'rV_n -> R] [K : R], (forall x : 'rV_n, `|F x| <= K) -> 0 <= K
Statement as written in the source
Local Lemma herbst_K0 : 0 <= K.
Used by 7 declarations in 1 file
forall {R : realType} {n : nat} [DF : 'I_n -> 'rV_n -> R] [L : R], (forall x : 'rV_n, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> forall (j : 'I_n) (x : 'rV_n), `|DF j x| <= `|L|
Statement as written in the source
Local Lemma herbst_DFL j x : `|DF j x| <= `|L|.
Used by 4 declarations in 1 file
- this file: herbst_df_bound, herbst_k_bound, herbst_ibp, herbst_df_lower
forall {R : realType} {n : nat} [DF : 'I_n -> 'rV_n -> R] [L : R], (forall x : 'rV_n, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> forall x y : 'rV_n, \sum_j DF j x * DF j y <= L ^+ 2
Statement as written in the source
Local Lemma herbst_DFdot x y : \sum_j DF j x * DF j y <= L ^+ 2.
Used by 1 declaration in 1 file
- this file: herbst_df_lower
forall {R : realType} {n : nat}, R -> 'rV_n * 'rV_n -> 'rV_nLocal Definition herbst_z (th : R) (p : 'rV[R]_n * 'rV[R]_n) : 'rV[R]_n := cos th *: p.1 + sin th *: p.2.
The interpolation between x and y along a quarter circle, and its derivative.
Used by 9 declarations in 1 file
forall {R : realType} {n : nat}, R -> 'rV_n * 'rV_n -> 'rV_nLocal Definition herbst_dz (th : R) (p : 'rV[R]_n * 'rV[R]_n) : 'rV[R]_n := - sin th *: p.1 + cos th *: p.2.
Used by 7 declarations in 1 file
forall {R : realType} {n : nat} (th : R) (p : 'rV_n * 'rV_n), is_derive th 1 (gauss_concentration.herbst_z^~ p) (gauss_concentration.herbst_dz th p)
Statement as written in the source
Used by 1 declaration in 1 file
- this file: herbst_is_derive_Fz
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> forall (th : R) (p : 'rV_n * 'rV_n), is_derive th 1 (fun th0 : R => F (gauss_concentration.herbst_z th0 p)) (\sum_j gauss_concentration.herbst_dz th p 0 j * DF j (gauss_concentration.herbst_z th p))
Statement as written in the source
Used by 1 declaration in 1 file
- this file: herbst_is_derive_f
forall {R : realType} {n : nat} (th : R), measurable_fun [set: 'rV_n * 'rV_n] (gauss_concentration.herbst_z th)
Statement as written in the source
Local Lemma herbst_mz (th : R) : measurable_fun [set: 'rV[R]_n * 'rV[R]_n] (herbst_z th).
Used by 2 declarations in 1 file
- this file: herbst_mf, herbst_mdf
forall {R : realType} {n : nat} (th : R), measurable_fun [set: 'rV_n * 'rV_n] (gauss_concentration.herbst_dz th)
Statement as written in the source
Local Lemma herbst_mdz (th : R) : measurable_fun [set: 'rV[R]_n * 'rV[R]_n] (herbst_dz th).
Used by 1 declaration in 1 file
- this file: herbst_mdf
forall {R : realType} {n : nat} (th : R) (p : 'rV_n * 'rV_n) (j : 'I_n), `|gauss_concentration.herbst_dz th p 0 j| <= `|p.1| + `|p.2|
Statement as written in the source
Local Lemma herbst_dz_bound (th : R) p j : `|herbst_dz th p 0 j| <= `|p.1| + `|p.2|.
Used by 1 declaration in 1 file
- this file: herbst_df_bound
forall {R : realType} {n : nat} [c s : R] [x : 'rV_n], continuous (fun v : 'rV_n => c *: x - s *: v)
Statement as written in the source
Local Lemma continuous_affine_rV (c s : R) (x : 'rV[R]_n) : continuous (fun v : 'rV[R]_n => c *: x - s *: v).
Used by 1 declaration in 1 file
- this file: herbst_ibp
forall {R : realType} {n : nat} (c s : R), measurable_fun [set: 'rV_n * 'rV_n] (fun p : ('rV_n * 'rV_n) => c *: p.1 - s *: p.2)
Statement as written in the source
Local Lemma measurable_affine2 (c s : R) : measurable_fun [set: 'rV[R]_n * 'rV[R]_n] (fun p => c *: p.1 - s *: p.2).
Used by 2 declarations in 1 file
- this file: herbst_ibp, herbst_df_lower
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> forall (c s : R) (x v : 'rV_n) (j : 'I_n), is_derive v (delta_mx 0 j) (fun v0 : 'rV[R]_n => F (c *: x - s *: v0)) (- s * DF j (c *: x - s *: v))
Statement as written in the source
Local Lemma herbst_is_derive_Fr (c s : R) (x v : 'rV[R]_n) j : is_derive v (delta_mx 0 j) (fun v => F (c *: x - s *: v)) (- s * DF j (c *: x - s *: v)).
Used by 1 declaration in 1 file
- this file: herbst_ibp
Variable t : R. Hypothesis t0 : 0 <= t.
forall {R : realType} {n : nat}, ('rV_n -> R) -> R -> R -> 'rV_n * 'rV_n -> RLocal Definition herbst_f (th : R) (p : 'rV[R]_n * 'rV[R]_n) : R := F p.1 * expR (t * F (herbst_z th p)).
Used by 5 declarations in 1 file
- this file: herbst_is_derive_f, herbst_mf, herbst_f_bound, herbst_phi_derive, gauss_herbst_ineq
forall {R : realType} {n : nat},
('rV_n -> R) -> ('I_n -> 'rV_n -> R) -> R -> R -> 'rV_n * 'rV_n -> RLocal Definition herbst_df (th : R) (p : 'rV[R]_n * 'rV[R]_n) : R := F p.1 * (expR (t * F (herbst_z th p)) * (t * \sum_j herbst_dz th p 0 j * DF j (herbst_z th p))).
Used by 7 declarations in 1 file
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> forall (t th : R) (p : 'rV_n * 'rV_n), is_derive th 1 ((gauss_concentration.herbst_f F t)^~ p) (gauss_concentration.herbst_df F DF t th p)
Statement as written in the source
Used by 1 declaration in 1 file
- this file: herbst_phi_derive
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> forall t th : R, measurable_fun [set: 'rV_n * 'rV_n] (gauss_concentration.herbst_f F t th)
Statement as written in the source
Local Lemma herbst_mf (th : R) : measurable_fun [set: 'rV[R]_n * 'rV[R]_n] (herbst_f th).
Used by 1 declaration in 1 file
- this file: herbst_phi_derive
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> forall t th : R, measurable_fun [set: 'rV_n * 'rV_n] (gauss_concentration.herbst_df F DF t th)
Statement as written in the source
Local Lemma herbst_mdf (th : R) : measurable_fun [set: 'rV[R]_n * 'rV[R]_n] (herbst_df th).
Used by 1 declaration in 1 file
- this file: integrable_herbst_df
forall {R : realType} {n : nat} [F : 'rV_n -> R] [K : R], (forall x : 'rV_n, `|F x| <= K) -> forall [t : R], 0 <= t -> forall x : 'rV_n, expR (t * F x) <= expR (t * K)
Statement as written in the source
Local Lemma herbst_expR_le (x : 'rV[R]_n) : expR (t * F x) <= expR (t * K).
Used by 5 declarations in 1 file
- this file: herbst_f_bound, herbst_df_bound, herbst_k_bound, herbst_df_lower, gauss_herbst_ineq
forall {R : realType} {n : nat} [F : 'rV_n -> R] [K : R], (forall x : 'rV_n, `|F x| <= K) -> forall [t : R], 0 <= t -> forall (th : R) (p : 'rV_n * 'rV_n), `|gauss_concentration.herbst_f F t th p| <= K * expR (t * K)
Statement as written in the source
Local Lemma herbst_f_bound (th : R) p : `|herbst_f th p| <= K * expR (t * K).
Used by 1 declaration in 1 file
- this file: herbst_phi_derive
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R] [K L : R], (forall x : 'rV_n, `|F x| <= K) -> (forall x : 'rV_n, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> forall [t : R], 0 <= t -> forall (th : R) (p : 'rV_n * 'rV_n), `|gauss_concentration.herbst_df F DF t th p| <= K * expR (t * K) * (t * (n%:R * `|L|)) * (`|p.1| + `|p.2|)
Statement as written in the source
Local Lemma herbst_df_bound (th : R) p : `|herbst_df th p| <= K * expR (t * K) * (t * (n%:R * `|L|)) * (`|p.1| + `|p.2|).
Used by 2 declarations in 1 file
- this file: integrable_herbst_df, herbst_phi_derive
forall {R : realType} {n : nat}, R -> R -> R -> 'rV_n * 'rV_n -> RLocal Definition herbst_D (p : 'rV[R]_n * 'rV[R]_n) : R := K * expR (t * K) * (t * (n%:R * `|L|)) * (`|p.1| + `|p.2|).
Used by 2 declarations in 1 file
- this file: herbst_D_ge0, integrable_herbst_D
forall {R : realType} {n : nat} [F : 'rV_n -> R] [K : R] (L : R), (forall x : 'rV_n, `|F x| <= K) -> forall [t : R], 0 <= t -> forall p : 'rV_n * 'rV_n, 0 <= gauss_concentration.herbst_D K L t p
Statement as written in the source
Local Lemma herbst_D_ge0 p : 0 <= herbst_D p.
Used by 2 declarations in 1 file
- this file: integrable_herbst_df, herbst_phi_derive
forall {R : realType} {n : nat} (K L t : R), (gauss n \x gauss n)%E.-integrable [set: 'rV_n * 'rV_n] (EFin \o gauss_concentration.herbst_D K L t)
Statement as written in the source
Local Lemma integrable_herbst_D : P.-integrable [set: 'rV[R]_n * 'rV[R]_n] (EFin \o herbst_D).
Used by 2 declarations in 1 file
- this file: integrable_herbst_df, herbst_phi_derive
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R] [K L : R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> (forall x : 'rV_n, `|F x| <= K) -> (forall x : 'rV_n, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> forall [t : R], 0 <= t -> forall th : R, (gauss n \x gauss n)%E.-integrable [set: 'rV_n * 'rV_n] (EFin \o gauss_concentration.herbst_df F DF t th)
Statement as written in the source
Local Lemma integrable_herbst_df (th : R) : P.-integrable [set: 'rV[R]_n * 'rV[R]_n] (EFin \o herbst_df th).
Used by 1 declaration in 1 file
- this file: herbst_df_lower
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R] [K L : R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> (forall x : 'rV_n, `|F x| <= K) -> (forall x : 'rV_n, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> forall [t : R], 0 <= t -> forall [th : R], th \in `]-1, 2[ -> is_derive th 1 (fun th0 : R => \int[(gauss n \x gauss n)%E]_p gauss_concentration.herbst_f F t th0 p) (\int[(gauss n \x gauss n)%E]_p gauss_concentration.herbst_df F DF t th p)
Statement as written in the source
Used by 1 declaration in 1 file
- this file: gauss_herbst_ineq
forall {R : realType} {n : nat},
('rV_n -> R) -> ('I_n -> 'rV_n -> R) -> R -> 'I_n -> 'rV_n -> RLocal Definition herbst_k (j : 'I_n) (x : 'rV[R]_n) : R := t * expR (t * F x) * DF j x.
Used by 4 declarations in 1 file
- this file: herbst_k_bound, herbst_mk, herbst_ibp, herbst_df_lower
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R] [K L : R], (forall x : 'rV_n, `|F x| <= K) -> (forall x : 'rV_n, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> forall [t : R], 0 <= t -> forall (j : 'I_n) (x : 'rV_n), `|gauss_concentration.herbst_k F DF t j x| <= t * expR (t * K) * `|L|
Statement as written in the source
Local Lemma herbst_k_bound j x : `|herbst_k j x| <= t * expR (t * K) * `|L|.
Used by 2 declarations in 1 file
- this file: herbst_ibp, herbst_df_lower
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> forall (t : R) (j : 'I_n), measurable_fun [set: 'rV_n] (gauss_concentration.herbst_k F DF t j)
Statement as written in the source
Local Lemma herbst_mk j : measurable_fun [set: 'rV[R]_n] (herbst_k j).
Used by 2 declarations in 1 file
- this file: herbst_ibp, herbst_df_lower
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R] [K L : R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> (forall x : 'rV_n, `|F x| <= K) -> (forall x : 'rV_n, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> forall [t : R], 0 <= t -> forall (c s : R) (j : 'I_n), \int[(gauss n \x gauss n)%E]_p (p.2 0 j * (gauss_concentration.herbst_k F DF t j p.1 * F (c *: p.1 - s *: p.2))) = \int[(gauss n \x gauss n)%E]_p (gauss_concentration.herbst_k F DF t j p.1 * (- s * DF j (c *: p.1 - s *: p.2)))
Gaussian integration by parts in the second component, for the test functions of the rotated covariance identity.
Statement as written in the source
Used by 1 declaration in 1 file
- this file: herbst_df_lower
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R] [K L : R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> (forall x : 'rV_n, `|F x| <= K) -> (forall x : 'rV_n, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> forall [t : R], 0 <= t -> forall [th : R], 0 <= sin th -> - (sin th * t * L ^+ 2) * \int[gauss n]_x expR (t * F x) <= \int[(gauss n \x gauss n)%E]_p gauss_concentration.herbst_df F DF t th p
Statement as written in the source
Used by 1 declaration in 1 file
- this file: gauss_herbst_ineq
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R] [K L : R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> (forall x : 'rV_n, `|F x| <= K) -> (forall x : 'rV_n, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> forall [t : R], 0 <= t -> \int[gauss n]_x (F x * expR (t * F x)) <= (\int[gauss n]_x F x + t * L ^+ 2) * \int[gauss n]_x expR (t * F x)
The covariance inequality behind Herbst's argument: E[F expR (t F)] <= (E[F] + t L^2) E[expR (t F)].
For x of law gauss n, F bounded of class C^1 with gradient norm at most L, and t >= 0: E[F expR (t F)] <= (E[F] + t L^2) E[expR (t F)].
Statement as written in the source
Used by 1 declaration in 1 file
- this file: gauss_concentration_smooth_ge0
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R] [K : R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> (forall x : 'rV_n, `|F x| <= K) -> forall s : R, is_derive s 1 (fun s0 : R => \int[gauss n]_x expR (s0 * F x)) (\int[gauss n]_x (F x * expR (s * F x)))
Statement as written in the source
Used by 1 declaration in 1 file
- this file: gauss_concentration_smooth_ge0
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R] [K L : R], (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> (forall x : 'rV_n, `|F x| <= K) -> (forall x : 'rV_n, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> forall [t : R], 0 <= t -> \int[gauss n]_x expR (t * (F x - \int[gauss n]_y F y)) <= expR (t ^+ 2 * L ^+ 2 / 2)
Statement as written in the source
Used by 1 declaration in 1 file
- this file: gauss_concentration_smooth
Context {R : realType} {n : nat}.
forall {R : realType} {n : nat} [F : 'rV_n -> R] [DF : 'I_n -> 'rV_n -> R] [K L : R] (t : R), (forall (j : 'I_n) (x : 'rV[R]_n), is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j : 'I_n, continuous (DF j)) -> (forall x : 'rV_n, `|F x| <= K) -> (forall x : 'rV_n, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> \int[gauss n]_x expR (t * (F x - \int[gauss n]_y F y)) <= expR (t ^+ 2 * L ^+ 2 / 2)
Gaussian concentration for a bounded function of class C^1 whose gradient has Euclidean norm at most L, with the sharp constant.
Statement as written in the source
Theorem gauss_concentration_smooth (F : 'rV[R]_n -> R) (DF : 'I_n -> 'rV[R]_n -> R) (K L t : R) : (forall j x, is_derive x (delta_mx 0 j) F (DF j x)) -> (forall j, continuous (DF j)) -> (forall x, `|F x| <= K) -> (forall x, mxdot (\row_j DF j x) (\row_j DF j x) <= L ^+ 2) -> \int[gauss n]_x expR (t * (F x - \int[gauss n]_y F y)) <= expR (t ^+ 2 * L ^+ 2 / 2).
Used by 1 declaration in 1 file
- this file: lip_approx_mgf
Context {R : realType}.
forall {R : realType}, R -> R -> RDefinition clip (k x : R) : R := Num.max (- k) (Num.min k x).
The clipping of x to [-k, k].
Used by 21 declarations in 3 files
- GaussianAnalysis.random_matrix.goe: clip_pow_lipschitz, clip_pow_err, goe_clip
- GaussianAnalysis.random_matrix.wigner: goe_stat_poly_clip, continuous_clip, semicircle_clip, clip_comp_lipschitz, polyfun_goe_stat_clip, goe_stat_clip_mean_cvg, wigner, clip3_bound, semicircle_near
- this file: clip_bound, clip_id, clip_abs, clip_lipschitz, lip_approx, clipL, clipK, lip_approx_mgf, lip_approx_dom
forall {R : realType} [k : R] (x : R), 0 <= k -> `|clip k x| <= k
Statement as written in the source
Lemma clip_bound (k x : R) : 0 <= k -> `|clip k x| <= k.
Used by 9 declarations in 3 files
- GaussianAnalysis.random_matrix.goe: clip_pow_lipschitz, clip_pow_err, goe_clip_bound
- GaussianAnalysis.random_matrix.wigner: clip_comp_lipschitz, polyfun_goe_stat_clip, goe_stat_clip_mean_cvg, clip3_bound
- this file: clip_abs, clipK
forall {R : realType} [k x : R], `|x| <= k -> clip k x = x
Statement as written in the source
Lemma clip_id (k x : R) : `|x| <= k -> clip k x = x.
Used by 5 declarations in 3 files
- GaussianAnalysis.random_matrix.goe: clip_pow_err
- GaussianAnalysis.random_matrix.wigner: semicircle_clip, wigner
- this file: clip_abs, lip_approx_cvg
forall {R : realType} [k : R] (x : R), 0 <= k -> `|clip k x| <= `|x|
Statement as written in the source
Lemma clip_abs (k x : R) : 0 <= k -> `|clip k x| <= `|x|.
Used by 1 declaration in 1 file
- this file: lip_approx_dom
Statement as written in the source
Used by 4 declarations in 3 files
- GaussianAnalysis.random_matrix.goe: clip_pow_lipschitz
- GaussianAnalysis.random_matrix.wigner: continuous_clip, clip_comp_lipschitz
- this file: clipL
Context {R : realType} {n : nat}. Variables (F : 'rV[R]_n -> R) (L : R). Hypothesis FL : forall x y, `|F x - F y| <= L * frobenius (x - y).
Let cF : continuous F := lipschitz_frobenius_continuous FL.
Used by 1 declaration in 1 file
- this file: mF
Let mF : measurable_fun [set: 'rV[R]_n] F := continuous_measurable_mx cF.
Used by 1 declaration in 1 file
- this file: gauss_concentration_upper_tail
Used by 4 declarations in 1 file
- this file: M0, lip_approx_dom, lip_approx_cvg, lip_approx_mean
Let M0 : 0 <= M.
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> forall x : 'rV_n, `|F x| <= `|F 0| + `|L| * frobenius x
Statement as written in the source
Lemma lipschitz_frobenius_bound x : `|F x| <= `|F 0| + `|L| * frobenius x.
Used by 2 declarations in 1 file
- this file: integrable_gauss_lipschitz, gauss_poincare_lipschitz
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> (gauss n).-integrable [set: 'rV_n] (EFin \o F)
Statement as written in the source
Lemma integrable_gauss_lipschitz : (gauss n).-integrable [set: 'rV[R]_n] (EFin \o F).
Used by 22 declarations in 9 files
- GaussianAnalysis.matrix.covering: gopnorm_mean, colmax_sep
- GaussianAnalysis.matrix.tensor_packing: sudakov_mx_series, tensor_packing
- GaussianAnalysis.random_matrix.goe: integrable_goe_opnorm
- GaussianAnalysis.random_matrix.goe_edge: integrable_goe_lambda_max, integrable_goe_lambda_maxN
- this file: lip_approx_mean, gauss_concentration_lower_tail
- GaussianAnalysis.comparison.escape: integrable_gwidth, integrable_escape_inf, integrable_minmax_sph, integrable_gauss_opnorm, gauss_proj_mean_ge
- GaussianAnalysis.comparison.cgmt: integrable_infsupX, integrable_infsupY, infsup_compare, cgmt_lower, cgmt_upper
- GaussianAnalysis.comparison.cone_escape: cone_escape_lower_tail
- GaussianAnalysis.comparison.sudakov: gauss_max_lower, sudakov_minoration
forall {R : realType} {n : nat}, ('rV_n -> R) -> nat -> 'rV_n -> R
Local Definition lip_approx (k : nat) : 'rV[R]_n -> R := gauss_smooth (k.+1%:R^-1) (fun x => clip k.+1%:R (F x)).
The approximations of F: clip it to [-(k + 1), k + 1], then smooth it at the scale 1 / (k + 1).
Used by 6 declarations in 1 file
- this file: lip_approx_mgf, lip_approx_dom, lip_approx_cvg, mA, lip_approx_mean, gauss_concentration
Used by 5 declarations in 1 file
- this file: lip_approx_mgf, lip_approx_dom, lip_approx_cvg, mA, gauss_concentration
Let clipK k x : `|clip k.+1%:R (F x)| <= k.+1%:R.
Used by 5 declarations in 1 file
- this file: lip_approx_mgf, lip_approx_dom, lip_approx_cvg, mA, gauss_concentration
Let ek0 (k : nat) : 0 < k.+1%:R^-1 :> R.
Used by 3 declarations in 1 file
- this file: lip_approx_mgf, lip_approx_dom, lip_approx_cvg
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> forall (k : nat) (t : R), \int[gauss n]_x expR (t * (gauss_concentration.lip_approx F k x - \int[gauss n]_y gauss_concentration.lip_approx F k y)) <= expR (t ^+ 2 * L ^+ 2 / 2)
Statement as written in the source
Local Lemma lip_approx_mgf k t : \int[gauss n]_x expR (t * (lip_approx k x - \int[gauss n]_y lip_approx k y)) <= expR (t ^+ 2 * L ^+ 2 / 2).
Used by 1 declaration in 1 file
- this file: gauss_concentration
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> forall (k : nat) (x : 'rV_n), `|gauss_concentration.lip_approx F k x| <= `|F x| + `|L| * \int[gauss n]_y frobenius y
Statement as written in the source
Local Lemma lip_approx_dom k x : `|lip_approx k x| <= `|F x| + `|L| * M.
Used by 1 declaration in 1 file
- this file: lip_approx_mean
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> forall [x : 'rV_n], gauss_concentration.lip_approx F x0 x @[x0 --> \oo] --> F x
Statement as written in the source
Local Lemma lip_approx_cvg x : lip_approx ^~ x @ \oo --> F x.
Used by 2 declarations in 1 file
- this file: lip_approx_mean, gauss_concentration
Let mA k : measurable_fun [set: 'rV[R]_n] (lip_approx k).
Used by 2 declarations in 1 file
- this file: lip_approx_mean, gauss_concentration
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> \int[gauss n]_x gauss_concentration.lip_approx F k x @[k --> \oo] --> \int[gauss n]_x F x
Statement as written in the source
Local Lemma lip_approx_mean : (fun k => \int[gauss n]_x lip_approx k x) @ \oo --> \int[gauss n]_x F x.
Used by 1 declaration in 1 file
- this file: gauss_concentration
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> forall t : R, (\int[gauss n]_x (expR (t * (F x - \int[gauss n]_y F y)))%:E <= (expR (t ^+ 2 * L ^+ 2 / 2))%:E)%E
Gaussian concentration for Lipschitz functions (Tsirelson, Ibragimov and Sudakov), with the sharp constant.
Gaussian concentration (Tsirelson, Ibragimov, Sudakov): if F is L-Lipschitz for the Euclidean norm, then E[expR (t (F x - E F))] <= expR (t^2 L^2 / 2) for every real t.
theories/ by: Gaussian/Lipschitz/Concentration Statement as written in the source
Used by 4 declarations in 2 files
- GaussianAnalysis.random_matrix.goe: goe_opnorm_mgf
- this file: integrable_gauss_concentration, gauss_concentration_upper_tail, gauss_poincare_lipschitz
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> forall t : R, (gauss n).-integrable [set: 'rV_n] (EFin \o (fun x : 'rV_n => expR (t * (F x - \int[gauss n]_y F y))))
Statement as written in the source
Used by 3 declarations in 2 files
- GaussianAnalysis.random_matrix.goe: integrable_goe_opnorm_expR, goe_opnorm_mgf
- this file: gauss_poincare_lipschitz
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> forall [u : R], 0 <= u -> (gauss n [set x | (\int[gauss n]_y F y + u <= F x)%R] <= (expR (- u ^+ 2 / (2 * L ^+ 2)))%:E)%E
For such F and u >= 0, P(F x >= E F + u) and P(F x <= E F - u) are at most expR (- u^2 / (2 L^2)), and P(|F x - E F| >= u) is at most twice that.
Statement as written in the source
Used by 5 declarations in 3 files
- GaussianAnalysis.random_matrix.goe_edge: goe_lambda_max_tail, goe_lambda_maxN_tail
- this file: gauss_concentration_lower_tail, gauss_concentration_abs_tail
- GaussianAnalysis.comparison.escape: gauss_opnorm_tail
Context {R : realType} {n : nat}. Variables (F : 'rV[R]_n -> R) (L : R). Hypothesis FL : forall x y, `|F x - F y| <= L * frobenius (x - y).
Let mF : measurable_fun [set: 'rV[R]_n] F := continuous_measurable_mx (lipschitz_frobenius_continuous FL).
Used by 1 declaration in 1 file
- this file: gauss_concentration_abs_tail
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> forall [u : R], 0 <= u -> (gauss n [set x | (F x <= \int[gauss n]_y F y - u)%R] <= (expR (- u ^+ 2 / (2 * L ^+ 2)))%:E)%E
Stated in the README together with gauss_concentration_upper_tail.
Statement as written in the source
Used by 3 declarations in 2 files
- this file: gauss_concentration_abs_tail
- GaussianAnalysis.comparison.escape: gordon_escape_tail, gordon_projection_tail
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> forall [u : R], 0 <= u -> (gauss n [set x | (u <= `|F x - \int[gauss n]_y F y|)%R] <= (2 * expR (- u ^+ 2 / (2 * L ^+ 2)))%:E)%E
Stated in the README together with gauss_concentration_upper_tail.
Statement as written in the source
Used by 3 declarations in 2 files
- GaussianAnalysis.random_matrix.wigner: goe_stat_conc
- GaussianAnalysis.comparison.cgmt: cgmt_escape_tail, cgmt_aux_tail
Two inequalities for the exponential, by monotonicity: 2 y <= expR y - expR (- y) for y >= 0, and 2 + x^2 <= expR x + expR (- x), that is cosh x >= 1 + x^2 / 2.
Context {R : realType}.
forall {R : realType} (z : R), is_derive z 1 (fun z0 : R => expR (- z0)) (- expR (- z))
Statement as written in the source
Lemma is_derive_expRN (z : R) : is_derive z 1 (fun z : R => expR (- z)) (- expR (- z)).
Used by 2 declarations in 1 file
- this file: expR_sinh_ge, expR_cosh_ge
forall {R : realType} [y : R], 0 <= y -> 2 * y <= expR y - expR (- y)
Stated in the README together with expR_cosh_ge.
Statement as written in the source
Lemma expR_sinh_ge (y : R) : 0 <= y -> 2 * y <= expR y - expR (- y).
Used by 1 declaration in 1 file
- this file: expR_cosh_ge
forall {R : realType} (x : R), 2 + x ^+ 2 <= expR x + expR (- x)
2 + x^2 <= expR x + expR (- x) for every real x, and 2 y <= expR y - expR (- y) for y >= 0.
Statement as written in the source
Lemma expR_cosh_ge (x : R) : 2 + x ^+ 2 <= expR x + expR (- x).
Used by 2 declarations in 2 files
- this file: gauss_poincare_lipschitz
- GaussianAnalysis.comparison.sudakov: gauss_shift_sym
The Gaussian Poincare inequality for Lipschitz functions: the variance of F x is at most L^2. For t > 0, cosh (t Y) >= 1 + t^2 Y^2 / 2 with Y = F x - E[F], and gauss_concentration bounds E[expR (t Y)] and E[expR (- t Y)]; so E[Y^2] <= L^2 expR (t^2 L^2 / 2), and t -> 0.
Context {R : realType} {n : nat}. Variables (F : 'rV[R]_n -> R) (L : R). Hypothesis FL : forall x y, `|F x - F y| <= L * frobenius (x - y).
forall {R : realType} {n : nat} [F : 'rV_n -> R] [L : R], (forall x y : 'rV_n, `|F x - F y| <= L * frobenius (x - y)) -> \int[gauss n]_x (F x - \int[gauss n]_y F y) ^+ 2 <= L ^+ 2
Gaussian Poincaré inequality for Lipschitz functions: if F is L-Lipschitz for the Euclidean norm and x has law gauss n, then E[(F x - E F)^2] <= L^2.
theories/ by: Gaussian/Lipschitz/Variance Statement as written in the source
Used by 2 declarations in 2 files
- GaussianAnalysis.random_matrix.goe: goe_tr_var
- GaussianAnalysis.comparison.escape: gauss_proj_mean_ge