Useful Inequalities {x2 > 0} v0.

39b · January 25, 2023 binomial max { n

kk
k (n−k+1)k
, k!
}≤ n
k

k
≤ nk! ≤ en k

k
; n k

nn
≤ kk (n−k) n−k .

k √ n n
n
≤ n for n ≥ k ≥ 0; √4πn (1 − 8n 1
) ≤ 2n 1
≤ √4πn (1 − 9n



 n
2  n
 n

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


n1 +n2
tn
k n


i=1 i=1 i=1 k1 k2
≤ k1 +k2
; k
≥ t k
for t ≥ 1.
√ nH(α)
1 1 1 π n
2
n n n G ≤ αn ≤ G for G = √ , H(x) = − log2 (xx (1−x)1−x ).
  
p p p
2
|xi + yi |p |xi |p |yi |p 2πnα(1−α)
P P P
Minkowski ≤ + for p ≥ 1.
i=1 i=1 i=1 n o
Pd n
en d
≤ min nd + 1, d , 2n


1/p  1/q i=0 i
for n ≥ d ≥ 1.
n n n

1 1
|xi |p |yi |q
P P P
Hölder |xi yi | ≤ for p, q > 1, + = 1. n
n
n 1

2 o
, 2nH(α) , 2n e−2n 2 −α
Pαn 1−α
p q
for α ∈ (0, 12 ).



i=1 i=1 i=1 i=0 i
≤ min 1−2α αn

Bernoulli (1 + x)r ≥ 1 + rx for x ≥ −1, r ∈ R \ (0, 1). Reverse for r ∈ [0, 1].
1
binary entropy 4x(1 − x) ≤ H(x) ≤ (4x(1 − x))1/ ln (4) for x ∈ (0, 1).
(1 + x)r ≤ 1−rx
for x ∈ [−1, r1 ), r ≥ 0.
(1 + x)r ≤ 1 + x
r for x ≥ 0, r ∈ [−1, 0]. n n

√ n n 1/(12n+1)

√ n n 1/12n

n n


x+1 Stirling e e
≤ 2πn e
e ≤ n! ≤ 2πn e
e ≤ en e
(1 + x)r ≤ 1 + r
(2 − 1)x for x ∈ [0, 1], r ∈ R \ (0, 1).
(1 + nx)n+1 ≥ (1 + (n + 1)x)n
P 2
for x ∈ R, n ∈ N.
q
x
n
xi )1/n ≤ 1 1
xi 2 ≤
Q P P
means min xi ≤ ≤( xi ≤ P i ≤ max xi
P −1
xi n n xi
(a + b)n ≤ an + nb(a + b)n−1 for a, b ≥ 0, n ∈ N.
2
p 1/p , w ≥ 0,


x n x n
P P
exponential ex ≥ 1 + ≥ 1 + x;


1+ ≥ ex 1 − xn


for n ≥ 1, |x| ≤ n. power means Mp ≤ Mq for p ≤ q, where Mp = i wi |xi | i i wi = 1.
n n
In the limit M0 = i |xi |wi , M−∞ = mini {xi }, M∞ = maxi {xi }.
Q
xn x x n+x/2 x ex n



n!
+1≤e ≤ 1+ n ; e ≥ n for x, n > 0.
xy
wi |xi |p wi |xi |q
P P
x x y
xy + y x > 1; xy > x


i
x+y
; e > 1+ y > e x+y for x, y > 0. Lehmer ≤ P i for p ≤ q, wi ≥ 0.
wi |xi |p−1 q−1
P
1 1+x i i wi |xi |
2−x
< xx < x2 − x + 1; e2x ≤ 1−x
for x ∈ (0, 1).
√ √ √  1
√ √ 2
x x+ y x−y x+ y x+y
x1/r (x − 1) ≤ rx(x1/r − 1) for x, r ≥ 1; 2−x ≤ 1 − 2
for x ∈ [0, 1]. log mean xy ≤ 2
(xy) 4 ≤ ln(x)−ln(y)
≤ 2
≤ 2
for x, y > 0.
3 2 2
x
xex ≥x+ x2 + 2
; ex ≤x+ ex ; ex + e−x ≤ 2ex /2 for x ∈ R.
√ x1−α y α +xα y 1−α x+y
e−x ≤ 1 − x2 for x ∈ [0, 1.59]; ex ≤ 1 + x + x2 for x < 1.79. Heinz xy ≤ 2
≤ 2
for x, y > 0, α ∈ [0, 1].

p
q
1+ x
p
≥ 1 + xq for (i) x > 0, p > q > 0,
Maclaurin- Sk 2 ≥ Sk−1 Sk+1 and (Sk )1/k ≥ (Sk+1 )1/(k+1) for 1 ≤ k < n,
(ii) − p < −q < x < 0, (iii) − q > −p > x > 0. Reverse for: 1 P
Newton Sk =  n ai 1 ai 2 · · · aik , and ai ≥ 0.
(iv) q < 0 < p , −q > x > 0, (v) q < 0 < p , −p < x < 0. k 1≤i1 <···<ik ≤n

P
P P
x x(6+x) Jensen ϕ pi xi ≤ i pi ϕ (xi ) where pi ≥ 0, pi = 1, and ϕ convex.
logarithm 1+x
≤ ln(1 + x) ≤ 6+4x
≤x for x > −1. i

2 1 ln(1+x) 2+x Alternatively: ϕ (E [X]) ≤ E [ϕ(X)]. For concave ϕ the reverse holds.
≤ √ ≤ ≤ √1 ≤ for x > −1.
2+x 1+x+x2 /12 x x+1 2+2x
n
P n
P n

P

Pn
ln(n) + 1
< ln(n + 1) < ln(n) + 1
≤ 1
≤ ln(n) + 1 for n ≥ 1. Chebyshev f (xi )g(xi )pi ≥ f (xi )pi g(xi )pi
n+1 n i=1 i i=1 i=1 i=1
1
y
P
1
|ln (x)| ≤ 2
|x − x1 |; ln(x + y) ≤ ln(x) + x
; ln(x) ≤ y(x y − 1); x, y > 0. for x1 ≤ · · · ≤ xn and f, g nondecreasing, pi ≥ 0, pi = 1.
     
x 2 Alternatively: E f (X)g(X) ≥ E f (X) E g(X) .
ln(1 + x) ≥ x − 2
for x ≥ 0; ln(1 + x) ≥ x − x2 for x ≥ −0.68.
n n n
x3 x cos x √ x
P P P
trigonometric x− 2
≤ x cos x ≤ 1−x2 /3
≤ x 3 cos x ≤ x − x3 /6 ≤ x cos √ ≤ sin x, rearrangement ai b i ≥ ai bπ(i) ≥ ai bn−i+1 for a1 ≤ · · · ≤ an ,
3 i=1 i=1 i=1
3 2
x 2 x b1 ≤ · · · ≤ bn and π a permutation of [n]. More generally:
hyperbolic x cos x ≤ sinh 2 x ≤ x cos (x/2) ≤ sin x ≤ (x cos x + 2x)/3 ≤ sinh x ,
2 n n n
−x2 2
n o
max π2 , π ≤ sinx x ≤ cos x2 ≤ 1 ≤ 1 + x3 ≤ tan x π
  P P P
π 2 +x2 x
for x ∈ 0, 2
. fi (bi ) ≥ fi (bπ(i) ) ≥ fi (bn−i+1 )
i=1 i=1 i=1
√ √ √ √ √ √

square root 2 x+1−2 x < √1 < x+1− x−1 < 2 x−2 x−1 for x ≥ 1. with fi+1 (x) − fi (x) nondecreasing for all 1 ≤ i < n.
x
n n n
√
Q Q Q
x+1 (x−1)2 x+1 (x−1)2 Dually: (ai + bi ) ≤ (ai + bπ(i) ) ≤ (ai + bn−i+1 ) for ai , bi ≥ 0.
x≤ 2
− 2
≤ x≤ 2
− 8
for x ∈ [0, 1]. i=1 i=1 i=1
 P Q
wi
Pn 
Weierstrass 1− w i xi ≤ and 1 − xi , Pn ai bi Pn
Pn

Milne i=1 (ai + bi ) ≤ ai i=1 bi for ai , bi ≥ 0.
i i

−wi
wi i=1 ai +bi i=1
P Q Q
1 + i w i x i ≤ i 1 + xi ≤ i 1 − xi for xi ∈ [0, 1], wi ≥ 1.
Q
wi P
−1 P Pn Q 1/k
k Pn
i 1 ± xi ≤ 1 ∓ i w i xi if additionally i wi xi ≤ 1. Carleman k=1 i=1 |ai | ≤e k=1 |ak |
 2 P
2
2 2 A
P
P

Kantorovich xi yi ≤ xi yi
for xi , yi > 0, Q n
ai −
Qn  Pn
i=1 bi
i i G i sum & product  ≤ |ai − bi | for |ai |, |bi | ≤ 1.
i=1 i=1
xi √
0<m≤ ≤ M < ∞, A = (m + M )/2, G = mM . Qn
(t + ai ) ≥ (t + 1)n where n
Q
yi i=1 i=1 ai ≥ 1, ai > 0, t > 0.
Pn ai n Pn p p
Nesbitt ≥ for ai ≥ 0, S = ai . ( xi )
P
i S−ai i=1 xi i
n−1
P
Radon i ap−1 ≥ p−1 for xi , ai ≥ 0, p ≥ 1 (rev. if p ∈ [0, 1]).
( i ai )
P
i
RU PU R U +1
sum & integral L−1 f (x) dx ≤ i=L f (i) ≤ L f (x) dx for f nondecreasing. Karamata
Pn
ϕ(ai ) ≥
Pn
ϕ(bi )
for a1 ≥ a2 ≥ · · · ≥ an , b1 ≥ · · · ≥ bn ,
i=1 i=1
and {ai }  {bi } (majorization), i.e. ti=1 ai ≥ ti=1 bi for all 1 ≤ t ≤ n,
P P
f (b)−f (a)
Cauchy f ′ (a) ≤ ≤ f ′ (b) where a < b, and f convex. Pn Pn
b−a with i=1 ai = i=1 bi , and ϕ convex (for concave ϕ the reverse holds).

xa · · · xa xbπ(1) · · · xbπ(n)
 
a+b 1
Rb ϕ(a)+ϕ(b) 1 1
P n
P n
Hermite ϕ ≤ ϕ(x) dx ≤ for ϕ convex. Muirhead π π(1) π(n)
≥ π , sums over permut. π of [n],
2 b−a a 2
where a1 ≥ · · · ≥ an , b 1 ≥ · · · ≥ bn , {ak }  {bk }, xi ≥ 0.
P ai a
Gibbs i ai log bi
≥ a log b
for ai , bi ≥ 0, or more generally:
P∞ P∞ am bn P∞
1 P∞
1
P bi
b

P P Hilbert m=1 n=1 m+n ≤π m=1 a2m 2 2 2
n=1 bn for am , bn ∈ R.
i ai ϕ ai
≤aϕ a
for ϕ concave, and a = ai , b = bi .
With max{m, n} instead of m + n, we have 4 instead of π.
n n n
ai
ai a i ≥ ai aπ(i)
P Q Q
Chong aπ(i)
≥n and for ai > 0. P∞  a1 +a2 +···+an p  p p P∞ p
i=1 i=1 i=1 Hardy n=1 n
≤ p−1 n=1 an for an ≥ 0, p > 1.

Schur-Vornicu f (x)(x − y)k (x − z)k + f (y)(y − z)k (y − x)k + f (z)(z − x)k (z − y)k ≥ 0 1 P∞ 2n 1
Mathieu c2 +1/2
< n=1 (n2 +c2 )2 < c2
for c 6= 0.
where x, y, z ≥ 0, k ≥ 1 integer, f convex or monotonic, f ≥ 0.
p q
Young ( px1p + qy1q )−1 ≤ xy ≤ xp + yq for x, y, p, q > 0, p1 + 1q = 1. 2−c(i) ≤ 1
P
Kraft for c(i) depth of leaf i of binary tree, sum over all leaves.
Ra R b −1
0 f (x) dx + 0 f (x) dx ≥ ab, for f cont., strictly increasing.
n
−1
≤ 1, A ⊂ 2[n] , no set in A is subset of another set in A.
P
LYM |X|
n X∈A
P xi n
Shapiro xi+1 +xi+2
≥ 2
where xi > 0, (xn+1 , xn+2 ) := (x1 , x2 ),
i=1
and n ≤ 12 if even, n ≤ 23 if odd. FKG Pr[x ∈ A ∩ B] ≥ Pr[x ∈ A] · Pr[x ∈ B], for A, B monotone set systems.

n P
n
|A|t ≤ for A, F ⊆ 2[n] , where every i ∈ [n]
Q
(det A)2 ≤ A2ij Shearer |traceF (A)|
Q
Hadamard where A is an n × n matrix. F ∈F
i=1 j=1
appears in at least t sets of F , and traceF (A) = {F ∩ A : A ∈ A}.
Pn 2
Pn Pn 2
Pk Pk
Schur i=1 λi ≤ i=1 j=1 Aij and i=1 di ≤ i=1 λi for 1 ≤ k ≤ n. vc(A)
n

≤ nvc(A) + 1 for A ⊆ 2[n] , and
P
A is an n × n matrix. For the second inequality A is symmetric. Sauer-Shelah |A| ≤ |str(A)| ≤ i
i=0
λ1 ≥ · · · ≥ λn the eigenvalues, d1 ≥ · · · ≥ dn the diagonal elements. str(A) = {X ⊆ [n] : traceX (A) = 2X }, vc(A) = max{|X| : X ∈ str(A)}.
Qn a Pn
x i a i xi qP qP
Qn i=1 i a ≤ Pn i=1 for xi ∈ [0, 12 ], ai ∈ [0, 1],  ≥ √1
P
2 2

ai = 1.
P
Ky Fan Khintchine i ai ≥ E

i ai ri i ai where ai ∈ R, and
i=1 (1 − xi ) i=1 ai (1 − xi )
i 2
1
Pn
2 Pn Pn ri ∈ {±1} random variables (r.v.) i.i.d. w.pr. 2
.
≥ a21 − 2
b21 2



Aczél a1 b 1 − i=2 ai b i i=2 ai − i=2 bi
Pn Pn n k
given that a21 > i=2 a2i or b21 > 2
i=2 bi . Bonferroni Pr
 W 
Ai ≤
P
(−1)j−1 Sj for 1 ≤ k ≤ n, k odd (rev. for k even),
n n n i=1 j=1
Q
1/n Q 1/n Q 1/n
Mahler xi + yi ≥ xi + yi where xi , yi > 0. Sk =
P 
Pr Ai1 ∧ · · · ∧ Aik

where Ai are events.
i=1 i=1 i=1
1≤i1 <···<ik ≤n
k
P n
P k
P


Abel b1 · min ai ≤ ai bi ≤ b1 · max ai for b1 ≥ · · · ≥ bn ≥ 0. Bhatia-Davis Var[X] ≤ M − E[X] E[X] − m where X ∈ [m, M ].
k i=1 i=1 k i=1
 √ √
Samuelson µ − σ n − 1 ≤ xi ≤ µ + σ n − 1 for i = 1, . . . , n,   Var[X]
Paley-Zygmund Pr X ≥ µ E[X] ≥ 1 − for X ≥ 0,
where µ =
P
xi /n , σ 2 =
P
(xi − µ)2 /n. (1 − µ)2 (E[X])2 + Var[X]
    Var[X] < ∞, and µ ∈ (0, 1).
Markov Pr |X| ≥ a ≤ E |X| /a where X is a r.v., a > 0. q
Pr X − E[X] ≥ λσ ≤ 9λ4 2 8
  
if λ ≥ ,
   
Pr X ≤ c ≤ (1 − E[X])/(1 − c) for X ∈ [0, 1] and c ∈ 0, E[X] . Vysochanskij- 3
 2
Pr X ∈ S] ≤ E[f (X)]/s for f ≥ 0, and f (x) ≥ s > 0 for all x ∈ S. 4τ 2τ
  
Petunin-Gauss Pr X − m ≥ ε ≤ 9ε2 if ε ≥ √ ,
3
ε 2τ

Pr X − E[X] ≥ t ≤ Var[X]/t2 where t > 0.
    
Chebyshev Pr X − m ≥ ε ≤ 1 − √
  if ε ≤ √ .
3τ 3
Pr X − E[X] ≥ t ≤ Var[X]/(Var[X] + t2 ) where t > 0.
 
Where X is a unimodal r.v. with mode m,
2nd moment Pr X > 0 ≥ (E[X])2 /(E[X 2 ]) where E[X] ≥ 0. σ 2 = Var[X] < ∞, τ 2 = Var[X] + (E[X] − m)2 = E[(X − m)2 ].
 

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

     

Etemadi Pr max |Sk | ≥ 3α ≤ 3 max Pr |Sk | ≥ α
1≤k≤n 1≤k≤n
E (X − µ)k
 
where Xi are i.r.v., Sk = ki=1 Xi , α ≥ 0.
P
kth moment
  
Pr X − µ ≥ t ≤ and
tk    
  

nk
k/2 Doob Pr max1≤k≤n |Xk | ≥ ε ≤ E |Xn | /ε for martingale (Xk ) and ε > 0.
Pr X − µ ≥ t ≤ Ck for Xi ∈ [0, 1] k-wise indep. r.v.,
et2
√ n

nσ 2 Mε


Xi , i = 1, . . . , n, µ = E[X], Ck = 2 πke1/6k , k even.
P  P 
X= Bennett Pr Xi ≥ ε
≤ exp − 2 θ where Xi i.r.v.,
M nσ 2
i=1
 
3/2 1
E X2 E[Xi ] = 0, σ 2 = n
P
Var[Xi ], |Xi | ≤ M (w. prob. 1), ε ≥ 0,
4th moment where 0 < E X 4 < ∞.
   
E |X| ≥
1/2
E [X 4 ] θ(u) = (1 + u) log(1 + u) − u.

Pr[X ≥ t] ≤ F (a)/at for X r.v., Pr[X = k] = pk , n −ε2

 
Chernoff  P 
Bernstein Pr Xi ≥ ε ≤ exp 2
for Xi i.r.v.,
F (z) = k pk z k probability gen. func., and a ≥ 1.
P
i=1 2(nσ + M ε/3)
1 P
Pr X ≥ ∆ = Pr X ≤ −∆ ≤ exp −∆2 /2n
   

for Xi ∈ {0, 1} i.r.v., E[Xi ] = 0, |Xi | < M (w. prob. 1) for all i, σ 2 = n Var[Xi ], ε ≥ 0.

and Pr[Xi = ±1] = 12 for i = 1, . . . , n, X =

P
Xi , ∆ ≥ 0. −δ 2
 
  
Azuma Pr Xn − X0  ≥ δ ≤ 2 exp for martingale (Xk ) s.t.
2 n
µ P 2
eδ −µδ 2 i=1 ci
  
 
Pr X ≥ (1 + δ)µ ≤ ≤ exp  
(1 + δ)(1+δ) 2+δ Xi − Xi−1  < ci (w. prob. 1), for i = 1, . . . , n, δ ≥ 0.
µ
e−δ −µδ 2
  
n
   
Pr X ≤ (1 − δ)µ ≤ ≤ exp Efron-Stein Var[Z] ≤ 1
E
P
Z − Z (i)

2
for Xi , Xi ′ ∈ X i.r.v.,
(1 − δ)(1−δ) 2 2
P i=1
for Xi ∈ [0, 1] i.r.v., X = Xi , µ = E[X], δ ≥ 0.
f : X n → R, Z = f (X1 , . . . , Xn ), Z (i) = f (X1 , . . . , Xi ′ , . . . , Xn ).
Further from the mean: Pr X ≥ R ≤ 2−R for R ≥ 2eµ (≈ 5.44µ).
 
−2δ 2
 
for Xi , Xi ′ ∈ X i.r.v.,
  
Xi ≥ t ≤ n

k t

McDiarmid Pr Z − E[Z] ≥ δ ≤ 2 exp Pn
P 
Pr p / k for Xi ∈ {0, 1} k-wise i.r.v., E[Xi ] = p. 2
i=1 ci
k
(1+δ)µ
Pr X ≥ (1 + δ)µ ≤ n pk̂ / Z, Z (i) as before, s.t. Z − Z (i)  ≤ ci for all i, and δ ≥ 0.
 

 
k̂ k̂
for Xi ∈ [0, 1] k-wise i.r.v.,
P µ
k ≥ k̂ = ⌈µδ/(1 − p)⌉, E[Xi ] = pi , X = Xi , µ = E[X], p = n , δ > 0.
 
V  ∆
Janson M ≤ Pr B i ≤ M exp where Pr[Bi ] ≤ ε for all i,
2 − 2ε
−2δ 2
 
  
Hoeffding Pr X − E[X] ≥ δ ≤ 2 exp for Xi i.r.v.,
Q P
Pn 2 M = (1 − Pr[Bi ]), ∆ = Pr[Bi ∧ Bj ].
i=1 (bi − ai ) i6=j,Bi ∼Bj
P
Xi ∈ [ai , bi ] (w. prob. 1), X = Xi , δ ≥ 0. V  Q Q
Lovász 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 λ ∈ R: (i,j)∈D
 2
λ (b − a)2

for xi ∈ [0, 1) for all i = 1, . . . , n and D the dependency graph.
E eλX ≤ exp
 
8 If each Bi mutually indep. of all other events, except at most d,
1
Pr max |Sk | ≥ ε ≤ ε2 Var[Sn ] = ε12
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 ε > 0.
P
c b a László Kozma · latest version: http://www.Lkozma.net/inequalities_cheat_sheet