P. 1
math

|Views: 4|Likes:
jensen3
jensen3

See more
See less

06/19/2015

pdf

text

original

# Tutorial 8: Jensen inequality

1

8. Jensen inequality
¯ Deﬁnition 64 Let a, b ∈ R, with a < b. Let φ : ]a, b[→ R be an R-valued function. We say that φ is a convex function, if and only if, for all x, y ∈]a, b[ and t ∈ [0, 1], we have: φ(tx + (1 − t)y) ≤ tφ(x) + (1 − t)φ(y) ¯ Exercise 1. Let a, b ∈ R, with a < b. Let φ : ]a, b[→ R be a map. 1. Show that φ : ]a, b[→ R is convex, if and only if for all x1 , . . . , xn in ]a, b[ and α1 , . . . , αn in R+ with α1 + . . . + αn = 1, n ≥ 1, we have: φ(α1 x1 + . . . + αn xn ) ≤ α1 φ(x1 ) + . . . αn φ(xn ) 2. Show that φ : ]a, b[→ R is convex, if and only if for all x, y, z with a < x < y < z < b we have: y−x z−y φ(x) + φ(z) φ(y) ≤ z−x z−x
www.probability.net

then φ is continuous. and u. t ∈ [0. b[→ R is convex. but that φ fails to be continuous on [0. Let φ : ]a. Show that if φ : ]a. b[→ R be convex. u . Let x0 ∈]a. Deﬁne φ : [0. 1]. Show that φ : ]a. 1]. b[. v ∈]a. ∀x. 1] → R by φ(0) = 1 and φ(x) = 0 for all x ∈]0. 6.probability. www. y. b[→ R is convex if and only if for all x. v[: φ(x) − φ(x0 ) φ(u ) − φ(u) φ(v ) − φ(v) ≤ ≤ u −u x − x0 v −v and deduce that limx↓↓x0 φ(x) = φ(x0 ) 5. v.Tutorial 8: Jensen inequality 2 3. Show that for all x ∈]x0 . b[ be such that u < u < x0 < v < v . z with a < x < y < z < b. 1]. y. Show that φ(tx + (1 − t)y) ≤ tφ(x) + (1 − t)φ(y).net . we have: φ(z) − φ(y) φ(y) − φ(x) ≤ y−x z−y 4.

probability. T ) be a topological space. . Let (Ω. 2. T ) be a topological space.net . . Exercise 2. T ) is compact. 1. there exists a ﬁnite subset {i1 . for all family (Vi )i∈I of open sets in Ω. . In short. We say that K is a compact subset of Ω. We say that (Ω. T ) be a topological space. in } of I such that Ω = Vi1 ∪ . ∪ Vin .Tutorial 8: Jensen inequality 3 Deﬁnition 65 Let (Ω. from any open covering of Ω. 3. it is a compact subset of itself. T ) is compact if and only if. such that Ω = ∪i∈I Vi . if and only if the induced topological space (K. . one can extract a ﬁnite sub-covering. we say that (Ω. then K is also a compact subset of Ω. T|K ) is a compact topological space. T ) is a compact topological space if and only if. and K ⊆ Ω. Show that ∅ is a compact subset of Ω. Show that if (Ω. . . Show that if Ω ⊆ Ω and K is a compact subset of Ω . Deﬁnition 66 Let (Ω. www.

www. 6. then K is a compact subset of Ω. 1. . . Let c = sup A.probability. Show that if (Vi )i∈I is a family of open sets in Ω such that K ⊆ ∪i∈I Vi .Tutorial 8: Jensen inequality 4 4. Let a. Show that there is > 0 such that a + ∈ A. if and only if for any family (Vi )i∈I of open sets in Ω such that K ⊆ ∪i∈I Vi . T ) is compact and K is closed in Ω. Let (Vi )i∈I be a family of open sets in R such that [a. 2. We deﬁne A as the set of all x ∈ [a. . b] such that [a. then K = ∪i∈I (Vi ∩ K) and Vi ∩ K is open in K for all i ∈ I. Show that if (Ω. . ∪ Vin . b ∈ R. Exercise 3. there is a ﬁnite subset {i1 . 5. b] ⊆ ∪i∈I Vi . . x] can be covered by a ﬁnite number of Vi ’s. Show that K ⊆ Ω is a compact subset of Ω. .net . Show that a ∈ A. in } of I such that K ⊆ Vi1 ∪ . a < b.

c with a < c < c < c . Show the existence of i0 ∈ I and c . 7. Show that [a. b ∈ R. Show that [a. Show that b ∧ c ≤ c and conclude that c = b.Tutorial 8: Jensen inequality 5 3.probability. b] is a compact subset of R. The closed interval [a. a < b. Theorem 34 Let a. 6. c ] ⊆ Vi0 .net . 8. b] is a compact subset of R. Show that a < c ≤ b. Show that [a. 5. www. such that ]c . c ] can be covered by a ﬁnite number of Vi ’s. 4. c ] can be covered by a ﬁnite number of Vi ’s.

Let (Ωi . Ti )i∈I be a family of Hausdorﬀ topological spaces.net . T ) be a topological space. T ) is a Hausdorﬀ topological space. www. then it is Hausdorﬀ. Let (Ω. ¯ 3. there exists open sets U and V in Ω. Let K be a compact subset of Ω and suppose there exists y ∈ K c . Show that if (Ω. then the induced topological space (Ω . such that: x∈U . y ∈ Ω with x = y. U ∩V =∅ Exercise 4. Exercise 5. T|Ω ) is itself Hausdorﬀ. y ∈V . T ) be a topological space. We say that (Ω. T ) is Hausdorﬀ and Ω ⊆ Ω.Tutorial 8: Jensen inequality 6 Deﬁnition 67 Let (Ω. T ) is metrizable. Show that if (Ω. Show that the product topological space Πi∈I Ωi is Hausdorﬀ. Show that any subset of R is Hausdorﬀ.probability. 1. if and only if for all x. 2. 4. Let (Ω. T ) be a Hausdorﬀ topological space.

. . . 2. x ∈ Wx and Vx ∩ Wx = ∅. Theorem 35 Let (Ω.probability. . if K is a compact subset.Tutorial 8: Jensen inequality 7 1. Show that K c = ∪y∈K c V y 6. Show that for all x ∈ K. Show that y ∈ V y ⊆ K c . . 4. For all K ⊆ Ω. T ) be a Hausdorﬀ topological space.net . . then it is closed. 5. www. Let V y = Vx1 ∩. Show that V y is open and V y ∩W y = ∅.∩Vxn . . 3. there are open sets Vx . Show that K is closed in Ω. such that y ∈ Vx . ∪ Wxn . xn } of K such that K ⊆ W y where W y = Wx1 ∪ . Wx in Ω. Show that there exists a ﬁnite subset {x1 . .

Let φ : R →]− 1. Show that δ(A) = 0 if and only if A = {x} for some x ∈ E. d) be a metric space. Deﬁne d (x. the element of R denoted δ(A). Deﬁnition 69 Let (E. y) = |φ(x) − φ(y)|.probability. Show that R is bounded with respect to d but not with respect to d . y) : x. 1[ be an increasing homeomorphism. 1.Tutorial 8: Jensen inequality 8 Deﬁnition 68 Let (E. y) = |x − y| and d (x. and A ⊆ E. For all A ⊆ E. www. deﬁned as δ(A) = sup{d(x.net . with the convention that δ(∅) = −∞. we ¯ call diameter of A with respect to d. y ∈ R. if and only if δ(A) < +∞. 2. Let (E. d) be a metric space. d) be a metric space. for all x. Show that d is a metric on R inducing the usual topology on R. Let A ⊆ E. We say that A is bounded. y ∈ A}. Exercise 6.

. . 4. . Suppose K is a closed subset of R which is bounded with respect to the usual metric on R. www. for all there is a ﬁnite subset {x1 . 4.Tutorial 8: Jensen inequality 9 3. is bounded with respect to any metric inducing the topology T . ) ∪ . 1. Show that K is a compact subset of [−M. Show that K is a compact subset of R.probability. ) > 0. Show that if K ⊆ E is a compact subset of E. M ]. . Show that K is also closed in [−M. M ]. xn } of K such that: K ⊆ B(x1 .net . . . M ]. 3. 2. Show that there exists M ∈ R+ such that K ⊆ [−M. ∪ B(xn . Show that any compact subset of any metrizable topological space (Ω. Exercise 7. T ).

www.Tutorial 8: Jensen inequality 10 5. then f (Ω) is a compact subset of (S. Let f : (Ω. TS ). 6. T ) → (S. TS ) be two topological spaces. TS ) be a continuous map. then the family (f −1 (Wi ))i∈I is an open covering of Ω.net . Show that any compact subset of R is closed and bounded.probability. Show that if (Ω. Show that if (Wi )i∈I is an open covering of f (Ω). and bounded with respect to the usual metric on R. Let (Ω. Show the following: Theorem 36 A subset of R is compact if and only if it is closed. 1. T ) is a compact topological space. T ) and (S. Exercise 8. 2.

¯ 3. Show that a subset of R is compact if and only if it is closed. then for all U ∈ TR with α ∈ U . then we have sup A ∈ A and inf A ∈ A. Let A be a non-empty subset of R. Show that if A is a non-empty closed subset of R. and let α = sup A. Show that if α = −∞. there exists ¯ ¯ β ∈ R with β < α and ]β. sin(x) = 0}. Conclude that α ∈ A. Show that if A is a non-empty. Show that any compact subset of R is a compact subset of R.net . 7. closed and bounded subset of R. TR ) is a compact topological space. then sup A ∈ A and inf A ∈ A. but that sup A ∈ A and inf A ∈ A. Show that A is closed in R. Show that (R. ¯ ¯ 1. ¯ 2. α] ⊆ U . www.probability.Tutorial 8: Jensen inequality 11 Exercise 9. 6. ¯ 5. Consider A = {x ∈ R . ¯ 4.

such that: f (xm ) = inf f (x) . Then.net . where (Ω. the continuity of f with respect to TR ¯ is equivalent to the continuity of f with respect to TR . T ) is a non-empty topological space. Let a. T ) is compact. T ) → (R. TR ) be a continuous map.Tutorial 8: Jensen inequality 12 Exercise 10. f attains its maximum and minimum. T ) be a compact. f (xM ) = sup f (x) x∈Ω x∈Ω Exercise 11.probability. with f (a) = f (b). b] → R be continuous on [a. and diﬀerentiable on ]a. 1. i. Let f : [a.e. xM ∈ Ω. ¯ ¯ Let f : (Ω. a < b. 2. Let (Ω. www. T ) → (R. Show the following: ¯ ¯ Theorem 37 Let f : (Ω. non-empty topological space. b]. there exist xm . Show that if f (Ω) ⊆ R. b ∈ R. TR ) be a continuous map. if (Ω. b[.

a < b. b[. b] → R be continuous on [a. b[ such that: f (b) − f (a) = (b − a)f (c) www. b[. Deﬁne: h(x) = f (x) − (x − a) f (b) − f (a) b−a 1. Then. with f (a) = f (b). Show the following: Theorem 38 (Rolle) Let a.b] f (x). Show that if c ∈]a. Show the existence of c ∈]a.net .Tutorial 8: Jensen inequality 13 1. 2. b[. Show that h is continuous on [a. b]. 2.probability. there exists c ∈]a. b] and diﬀerentiable on ]a. Let f : [a. b ∈ R. b ∈ R. Let a. b[ such that f (c) = 0. b[ and f (c) = supx∈[a. Let f : [a. a < b. b] and diﬀerentiable on ]a. b] → R be continuous on [a. and diﬀerentiable on ]a. Exercise 12. then f (c) = 0.

Let a. Show that h is continuous on [a. b]. a < b. 2. Deﬁne: n h(x) = f (b) − f (x) − k=1 (b − x)k (k) (b − x)n+1 f (x) − α k! (n + 1)! where α is chosen such that h(a) = 0. Let f : [a. We assume that f is of class C n on [a. Show that for all x ∈]a. b[: h (x) = 3. b[. and that f (n+1) exists on ]a.Tutorial 8: Jensen inequality 14 Exercise 13. Prove the following: (b − x)n (α − f (n+1) (x)) n! www. b] and diﬀerentiable on ]a. 1. b] → R be a map. b[. Let n ≥ 0.net . b ∈ R.probability.

www. b] such that f (n+1) exists on ]a. then for all x.net . we have: φ (x) ≤ φ (y) 2. Let a. Let f : [a. b[→ R be diﬀerentiable. a < b and φ : ]a. b] → R be a map of class C n on [a. with c1 < c2 and: φ(y) − φ(x) φ(z) − φ(y) = φ (c1 )(y − x) = φ (c2 )(z − y) 3. b[. b[ such that: n f (b) − f (a) = k=1 (b − a)k (k) (b − a)n+1 (n+1) f (a) + f (c) k! (n + 1)! ¯ Exercise 14. Then. c2 ∈]a. z ∈]a. there exists c ∈]a. then φ is convex. 1. b[.Tutorial 8: Jensen inequality 15 Theorem 39 (Taylor-Lagrange) Let a.probability. x < y. Show that if φ is convex. a < b. b[. b ∈ R. Show that if x. Show conversely that if φ is non-decreasing. there are c1 . b[ with x < y < z. y. b ∈ R. y ∈]a. and n ≥ 0.

and (S. Let X be a nonnegative random variable. Deﬁnition 71 Let (Ω. 5. P ) be a probability space. Σ) be a measurable space. F .r. F . F ) → (S. Show that x → ex is convex on R. P ) be a probability space.Tutorial 8: Jensen inequality 16 4. or an element of L1 (Ω. We call C expectation of X. to (S. if and only if P (Ω) = 1.probability. Σ). F .net . We call random variable w. the integral: E[X] = Ω XdP www. Show that x → − ln(x) is convex on ]0. F . any measurable map X : (Ω. +∞[. denoted E[X]. Σ). Deﬁnition 72 Let (Ω. Deﬁnition 70 we say that a ﬁnite measure space (Ω. P ) is a probability space. P ).

F ) → (R. Show that m ∈]a. Show that if E[X] = a. R 3. b[. 2. then b ∈ R and X = b P -a. 4. b[. Let m = E[X]. F . Show that φ ◦ X ∈ L1 (Ω.net Show that β ∈ R and that for all z ∈]m. Let a.s. b[. if and only if E[|φ ◦ X|] < +∞. B(R)) is measurable.probability.s. 1. 5. then a ∈ R and X = a P -a. b ∈ R.m[ φ(z) − φ(m) z−m www. Show that φ ◦ X : (Ω. we have: β≤ . P ). P ) be such R that X(Ω) ⊆]a. b[→ R be a convex map. F . 6. F . P ) be a probability space and X ∈ L1 (Ω. a < b and φ : ]a. Let (Ω.Tutorial 8: Jensen inequality 17 ¯ Exercise 15. Show that if E[X] = b. Deﬁne: β = sup φ(m) − φ(x) m−x x∈]a.

9. Then: R φ(E[X]) ≤ E[φ ◦ X] www. F . P ) is such that X(Ω) ⊆]a. Show that if φ ◦ X ∈ L1 (Ω. P ). b[→ R be a convex map. F . F . P ) then φ(m) ≤ E[φ ◦ X]. a < b and φ : ]a.net . φ(m) + β(X(ω) − m) ≤ φ(X(ω)).Tutorial 8: Jensen inequality 18 7. b[ and such R that φ ◦ X ∈ L1 (Ω. 8. F . R Theorem 40 (Jensen inequality) Let (Ω. Suppose that X ∈ L1 (Ω. Let a. we have φ(m) + β(x − m) ≤ φ(x). P ) be a probability ¯ space. b ∈ R. Show that for all x ∈]a.probability. b[. Show that for all ω ∈ Ω.

scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->