Useful Inequalities {x2 > 0} v0.

28 January 13, 2016 binomial max { n

kk
,
k (nk+1)k
k!
} n

k
nk
k!

(en)k
kk
and n

k
nn
kk (nk)nk
2n .
nk n n
n for n k 0 and 4n (1 8n 1
) 2n 1
4n (1 9n



 n
2  n
 n

4k! k n
).
x2i yi2
P P P
Cauchy-Schwarz xi y i n1 n2


n1 +n2

i=1 i=1 i=1
k1 k2
k +k for n1 k1 0, n2 k2 0.
1 2
1 1 1 nH()
n

n n n G n G for G = 2 , H(x) = log2 (xx (1x)1x ).
  
p p p
|xi + yi |p |xi |p |yi |p 2
P P P
Minkowski + for p 1. 2n(1)
i=1 i=1 i=1 Pd n
Pd n

i=0 i nd + 1 and i=0 i 2
n for n d 0.
n
 n
1/p  n
1/q
1 1 Pd n
en d
|xi |p |yi |q
P P P

Holder |xi yi | for p, q > 1, p
+ q
= 1. i=0 i d
for n d 1.
i=1 i=1 i=1 Pd n
n
d
for n


i=0 i d 1 + n2d+1 2
d 0.
Bernoulli (1 + x)r 1 + rx for x 1, r R \ (0, 1). Reverse for r [0, 1].
n n n 1 n
i=0 i 12 n for (0, 21 ).

P


(1 + x)r 1 + (2r 1)x for x [0, 1], r R \ (0, 1). n

1
(1 + x)n 1nx
for x [1, 0], n N. square root 2 x+12 x < 1 < x+1 x1 < 2 x2 x1 for x 1.
x
rx 1
(1 + x)r 1+ 1(r1)x
for x [1, r1
), r > 1.
n n

n n 1/(12n+1)

n n 1/12n

n n


(1 + nx)n+1 (1 + (n + 1)x)n for x R, n N. Stirling e e
2n e
e n! 2n e
e en e

(a + b)n an + nb(a + b)n1 for a, b 0, n N. q P 2

x
n
xi )1/n 1 1
xi 2

p
q Q P P
P i
1+ x 1 + xq for (i) x > 0, p > q > 0, means min xi P 1 ( n
xi n xi
max xi
p xi

(ii) p < q < x < 0, (iii) q > p > x > 0. Reverse for:
p 1/p , w 0,
P
P
(iv) q < 0 < p , q > x > 0, (v) q < 0 < p , p < x < 0. power means Mp Mq for p q, where Mp = i wi |xi | i i wi = 1.
In the limit M0 = i |xi |wi , M = mini {xi }, M = maxi {xi }.
Q
x n x n x2
ex 1 + ex 1




exponential n
1 + x, 1+ n n
for n > 1, |x| n.
wi |xi |p wi |xi |q
P P
i
ex xe for x R, and xn
+1 ex 1+ x n+x/2


for x, n > 0. Lehmer P i for p q, wi 0.
wi |xi |p1 q1
P
n! n i i wi |xi |
x2
ex 1 + x + 2
for x 0, reverse for x 0.   2
x+ y
 1 xy x+ y x+y
x x log mean xy (xy) 4 for x, y > 0.
ex 1 2
for x [0, 1.59] and 2x 1 2
for x [0, 1]. 2 ln(x)ln(y) 2 2

1
< xx < x2 x + 1 for x (0, 1).
2x x1 y +x y 1 x+y
Heinz xy 2
2
for x, y > 0, [0, 1].
x1/r (x 1) rx(x1/r
for x, r 1. 1)

y xy
xy + y x > 1 and ex > 1 + x
p
Sk 2 Sk1 Sk+1 k
p
(k+1)
y
> e x+y for x, y > 0. Maclaurin- and Sk Sk+1 for 1 k < n,
1 P
2 y exy 1 + x y + exy , and ex x + ex
2
for x, y R. Newton Sk =  n ai 1 ai 2 aik , and ai 0.
k 1i1 <<ik n

x1 x2 1 1
logarithm ln(x) x 1, ln(x) n(x n 1) for x, n > 0. P
P P
x 2x Jensen i pi xi i pi (xi ) where pi 0, pi = 1, and convex.
2x x
2+x
ln(1 + x) x+1
for x 0, reverse for x (1, 0]. Alternatively: (E [X]) E [(X)]. For concave the reverse holds.
1 Pn 1
ln(n + 1) < ln(n) + n
i=1 i ln(n) + 1 n
P n
P n

P
n
P
x
Chebyshev f (ai )g(bi )pi f (ai )pi g(bi )pi f (ai )g(bni+1 )pi
ln(1 + x) 2
for x [0, 2.51], reverse elsewhere. i=1 i=1 i=1 i=1
P
x2 x3 for a1 an , b1 bn and f, g nondecreasing, pi 0, pi = 1.
ln(1 + x) x 2
+ 4
for x [0, 0.45], reverse elsewhere.      
Alternatively: E f (X)g(X) E f (X) E g(X) .
x2 x3
ln(1 x) x 2
2
for x [0, 0.43], reverse elsewhere.
n n n

P P P
x3 x cos x x rearrangement ai b i ai b(i) ai bni+1 for a1 an ,
trigonometric x 2
x cos x 1x2 /3
x 3 cos x x x3 /6 x cos sin x, i=1 i=1 i=1
3
x3 x2 b1 bn and a permutation of [n]. More generally:
hyperbolic x cos x sinh2 x
x cos2 (x/2) sin x (x cos x + 2x)/3 sinh x
,
n
P n
P n
P
2
x sin x x cos(x/2) x x + x3
tan x

all for x 0,

. fi (bi ) fi (b(i) ) fi (bni+1 )
3 2 i=1 i=1 i=1
ex(+x/2)


cosh(x) + sinh(x) for x R, [1, 1]. with fi+1 (x) fi (x) nondecreasing for all 1 i < n.
 Q
wi P 1/k
Weierstrass 1 xi 1 w i xi where xi 1, and Pn Q
i i k Pn
Carleman k=1 i=1 |ai | e k=1 |ak |
either wi 1 (for all i) or wi 0 (for all i).
m Q
n n
m Q m P
n m P
n
P
If wi [0, 1], wi 1 and xi 1, the reverse holds. P P Q Q
sum & product aij ai(j) and aij ai(j)
j=1 i=1 j=1 i=1 j=1 i=1 j=1 i=1
xp yq
Young ( px1p + 1 1
qxq
) xy + for x, y 0, p, q > 0, 1
+ 1
= 1.
p q p q where 0 ai1 aim for i = 1, . . . , n and is a permutation of [n].
Q n Qn  Pn

 A 2 P i=1 ai i=1 bi i=1 |ai bi | for |ai |, |bi | 1.

2  
xi 2 yi 2 G
P
P
Kantorovich i i i xi yi for xi , yi > 0, Qn n
Q n
xi i=1 ( + ai ) (1 + ) , where i=1 ai 1, ai > 0, > 0.
0<m yi
M < , A = (m + M )/2, G = mM .
P  P  P  P 
1+x 1x 1x 1+x 1+y 1y 1y 1+y
RU PU R U +1 Callebaut i ai bi i ai bi i ai bi i ai bi
sum-integral L1 f (x) dx i=L f (i) L f (x) dx for f nondecreasing.
for 1 x y 0, and i = 1, . . . , n.
f (b)f (a) Pn Pn
Cauchy (a) ba
(b) where a < b, and convex. Karamata i=1 (ai ) i=1 (bi ) for a1 a2 an and b1 bn ,
Pt Pt
  and {ai }  {bi } (majorization), i.e. i=1 ai i=1 bi for all 1 t n,
a+b 1
Rb (a)+(b)
Hermite 2
ba a (x) dx 2
for convex. with equality for t = n and is convex (for concave the reverse holds).

1
xa 1
xa 1
xb(1)
1
xb(n)
P n
P n
n n n Muirhead n! (1) (n)
n!
ai
ai a i ai a(i)
P Q Q
Chong a(i)
n and for ai > 0.
i=1 i=1 i=1 where a1 a2 an and b1 b2 bn and {ak }  {bk },
P ai a xi 0 and the sums extend over all permutations of [n].
Gibbs i ai log bi
a log b
for ai , bi 0, or more generally:
bi
b

P P am bn P
1 P
1
a2m 2 2
P P P
i ai ai
a a
for concave, and a = ai , b = bi . Hilbert m=1 n=1 m+n m=1
2
n=1 bn for am , bn R.

n With max{m, n} instead of m + n, we have 4 instead of .

P xi n
Shapiro xi+1 +xi+2
2
where xi > 0, (xn+1 , xn+2 ) := (x1 , x2 ),
i=1 P  a1 +a2 ++an p  p p P p
Hardy n=1 n
p1 n=1 an for an 0, p > 1.
and n 12 if even, n 23 if odd.
P
4 P P
Schur xt (x y)(x z) + y t (y z)(y x) + z t (z x)(z y) 0 Carlson n=1 an 2 n=1 a2n n=1 n2 a2n for an R.
where x, y, z 0, t > 0
1 P 2n 1
n P
n
Mathieu c2 +1/2
< n=1 (n2 +c2 )2 < c2
for c 6= 0.
(det A)2 A2ij
Q
Hadamard where A is an n n matrix.
i=1 j=1
 P p
ak
pp an p
P P
Pn Pn Pk Pk Copson k
for an 0, p > 1, reverse if p (0, 1).
Schur i=1 2i i,j=1 A2ij and i=1 di i=1 i for 1 k n. n=1 kn n=1

A is an n n matrix. For the second inequality A is symmetric.

2c(i) 1
P
Kraft for c(i) depth of leaf i of binary tree, sum over all leaves.
1 n the eigenvalues, d1 dn the diagonal elements.
n
1
a 1, A 2[n] , no set in A is subset of another set in A.
Qn Pn P
x i a i xi LYM |X|
Qn i=1 i a Pn i=1 for xi [0, 12 ], ai [0, 1],
P
Ky Fan ai = 1. XA
i=1 (1 xi ) i=1 ai (1 xi )
i

Pn
2 Pn Pn vc(A)
n

a21 2 b2 2
for A 2[n] , and



a1 b 1 ai b i i=2 ai i=2 bi
P
Aczel i=2 1 Sauer-Shelah |A| |str(A)| i
Pn Pn i=0
given that a21 > 2
i=2 ai or b21 > 2
i=2 bi .
str(A) = {X [n] : X shattered by A}, vc(A) = max{|X| : X str(A)}.
n
1/n n n
Q Q 1/n Q 1/n
Mahler xi + yi xi + yi where xi , yi > 0. n k
(1)j1 Sj
 W 
i=1 i=1 i=1
P
Bonferroni Pr Ai for 1 k n, k odd,
i=1 j=1
k n k
P P P n k
Abel b1 min ai ai bi b1 max ai for b1 bn 0.
(1)j1 Sj
 W  P
k i=1 i=1 k i=1 Pr Ai for 2 k n, k even.
i=1 j=1
Pn
P n ai bi
 Pn
Pn P  
Sk = Pr Ai1 Aik where Ai are events.


Milne i=1 (ai + bi ) i=1 ai +bi i=1 ai i=1 bi
1i1 <<ik n
Bhatia-Davis


Var[X] M E[X] E[X] m


where X [m, M ].
  Var[X]
Paley-Zygmund Pr X E[X] 1 for X 0,
(1 )2 (E[X])2 + Var[X]
Var[X] < , and (0, 1).
Samuelson n 1 xi + n 1 for i = 1, . . . , n.
q
xi /n , 2 = (xi )2 /n.
P P
Where = Pr X E[X] 94 2 8
  
Vysochanskij- if 3
,
2
4 2
  
Petunin-Gauss Pr X m 92 if ,
    3
Markov Pr |X| a E |X| /a where X is a random variable (r.v.), a > 0.
2
  
    Pr X m 1
  if .
Pr X c (1 E[X])/(1 c) for X [0, 1] and c 0, E[X] . 3 3

Pr X S] E[f (X)]/s for f 0, and f (x) s > 0 for all x S. Where X is a unimodal r.v. with mode m,
2 = Var[X] < , 2 = Var[X] + (E[X] m)2 = E[(X m)2 ].
Pr X E[X] t Var[X]/t2 where t > 0.
  
Chebyshev    

Etemadi Pr max |Sk | 3 3 max Pr |Sk |
Pr X E[X] t Var[X]/(Var[X] + t2 ) where t > 0.
 
1kn 1kn

where Xi are i.r.v., Sk = ki=1 Xi , 0.

P
2nd moment Pr X > 0 (E[X])2 /(E[X 2 ]) where E[X] 0.
 

Pr X = 0 Var[X]/(E[X 2 ]) where E[X 2 ] 6= 0.

     
Doob Pr max1kn |Xk | E |Xn | / for martingale (Xk ) and > 0.

E (X )k
 
n n 2
 
kth moment M

  
Pr X  t and Bennett Pr
 P
Xi

exp 2 where Xi i.r.v.,
tk M n 2
 k/2 i=1
nk 1
E[Xi ] = 0, 2 = n
P
Var[Xi ], |Xi | M (w. prob. 1), 0,
  
Pr X  t Ck for Xi [0, 1] k-wise indep. r.v.,
t2
(u) = (1 + u) log(1 + u) u.
Xi , i = 1, . . . , n, = E[X], Ck = 2 ke1/6k 1.0004, k even.
P
X=
n 2
 
 P 
Chernoff Pr[X t] F (a)/at for X r.v., Pr[X = k] = pk , Bernstein Pr Xi exp 2
for Xi i.r.v.,
i=1 2(n + M /3)
F (z) = k pk z k probability gen. func., and a 1.
P
1 P
E[Xi ] = 0, |Xi | < M (w. prob. 1) for all i, 2 = n Var[Xi ], 0.

e 2
  
2
   
Pr X (1 + ) exp   
(1 + )(1+) 3 Azuma Pr Xn X0  2 exp for martingale (Xk ) s.t.
2 n 2
P
P i=1 ci
for Xi i.r.v. from [0,1], X = Xi , = E[X], 0 resp. [0, 1).  
Xi Xi1  < ci (w. prob. 1), for i = 1, . . . , n, 0.

e 2
  
 
Pr X (1 ) exp for [0, 1).  n

(1 )(1) 2 Efron-Stein Var[Z] 1
E
P
Z Z (i)

2
for Xi , Xi X i.r.v.,
2
Further from the mean: Pr X R 2R for R 2e ( 5.44). i=1
 
 
n
f : X n R, Z = f (X1 , . . . , Xn ), Z (i) = f (X1 , . . . , Xi , . . . , Xn ).
  k
pk P
Pr X t for Xi {0, 1} k-wise i.r.v., E[Xi ] = p, X = Xi .
2 2
 
t
 
for Xi , Xi X i.r.v.,
  
k McDiarmid Pr Z E[Z] 2 exp Pn 2
i=1 ci
  n
k (1+)

Pr X (1 + ) p / for Xi [0, 1] k-wise i.r.v., Z, Z (i) as before, s.t. Z Z (i)  ci for all i, and 0.
 
k k
P
k k = /(1 p), E[Xi ] = pi , X = Xi , = E[X], p = n , > 0.  
V 
Janson M Pr B i M exp where Pr[Bi ] for all i,
  

2 2
 2 2
Hoeffding Pr X E[X] 2 exp Pn 2
for Xi i.r.v., Q
M = (1 Pr[Bi ]), =
P
Pr[Bi Bj ].
i=1 (bi ai ) i6=j,Bi Bj
P
Xi [ai , bi ] (w. prob. 1), X = Xi , 0. V  Q Q
Lovasz Pr B i (1 xi ) > 0 where Pr[Bi ] xi (1 xj ),
A related lemma, assuming E[X] = 0, X [a, b] (w. prob. 1) and R: (i,j)D
 2
(b a)2

E eX exp
  for xi [0, 1) for all i = 1, . . . , n and D the dependency graph.
8 If each Bi mutually indep. of the set of all other events, exc. at most d,
Pr max |Sk | 12 Var[Sn ] = 12
V 
Pr[Bi ] p for all i = 1, . . . , n , then if ep(d + 1) 1 then Pr B i > 0.
  P
Kolmogorov Var[Xi ]
k i
where X1 , . . . , Xn are i.r.v., E[Xi ] = 0,
Var[Xi ] < for all i, Sk = ki=1 Xi and > 0.
P
Laszlo Kozma latest version: http://www.Lkozma.net/inequalities_cheat_sheet