Probability Distributions Cheat Sheet

PMF/PDF, CDF, Mean, Variance, MGF & Key Relations

Discrete Distributions

Discrete Uniform on {a, a + 1, . . . , b}, a ≤ b; N = b − a + 1.

1 ⌊k⌋ − a + 1
PMF: p(k) = for k = a, . . . , b. CDF: F (k) = for k ∈ [a, b] (stepwise).
N N
a+b 2
N −1
Mean: E[X] = . Var: Var(X) = .
2 12
ta
e e −1tN
MGF: MX (t) = (t ̸= 0), MX (0) = 1.
N et − 1
Notes: Finite support; as N → ∞ with scaling, approximates continuous uniform.

Bernoulli(p), p ∈ [0, 1].

PMF: p(k) = pk (1 − p)1−k , k ∈ {0, 1}. CDF: F (x) = 0 for x < 0; 1 − p for 0 ≤ x < 1; 1 for x ≥ 1.
Mean: E[X] = p. Var: Var(X) = p(1 − p). MGF: MX (t) = 1 − p + pet .
Notes: Building block of Binomial; indicator of event with prob p.

Binomial(n, p),
n ∈ N, p ∈ [0, 1].

PMF: p(k) = nk pk (1 − p)n−k , k = 0, . . . , n. CDF: Incomplete beta form.
Mean: np. Var: np(1 − p). MGF: (1 − p + pet )n .
Notes: Bin(n, p) → Poisson(λ) as n → ∞, p → 0 with np = λ (Poisson limit). Normal approx for large
n (CLT).

Poisson(λ), λ > 0.
λk
PMF: p(k) = e−λ , k = 0, 1, 2, . . . CDF: Incomplete gamma form.
k!
Mean/Var: E[X] = Var(X) = λ. MGF: exp λ(et − 1) .

Notes: Sum of independent Poissons is Poisson. Poisson–Gamma conjugacy (Bayes). Thinning &
superposition properties.

Geometric(p) (trials until first success), p ∈ (0, 1]; k = 1, 2, . . .

PMF: p(k) = p(1 − p)k−1 . CDF: F (k) = 1 − (1 − p)⌊k⌋ .
pet
Mean: 1/p. Var: (1 − p)/p2 . MGF: MX (t) = , valid for t < − ln(1 − p).
1 − (1 − p)et
Notes: Memoryless: Pr(X > m + n | X > m) = (1 − p)n . Alternative (failures before first success)
has support 0, 1, . . . .

Hypergeometric(N, K,
n): population N , K successes, draw n w/o replacement.

K N −K


k n−k
PMF: p(k) = N

, k = max(0, n − (N − K)), . . . , min(n, K).
n
K N −n
 
Mean: n K
N. Var: n K
N 1 − N N − 1. MGF: No simple closed form.
Notes: Without-replacement analog of Binomial. For large N with fixed p = K/N , ≈ Bin(n, p).

Negative Binomial(r,

r p) (trials until rth success), r > 0, p ∈ (0, 1]; k = r, r + 1, . . .
PMF: p(k) = k−1r−1 p (1 − p) k−r . CDF: Incomplete beta form.
 pet r
Mean: r/p. Var: r(1 − p)/p2 . MGF: MX (t) = , t < − ln(1 − p).
1 − (1 − p)et
Notes: If defined as failures before rth success (support 0, 1, . . . ), then E[X] = r 1−p 1−p
p , Var(X) = r p2 .
Sums of independent Geometrics ⇒ NegBin.

1
Continuous Distributions

Continuous Uniform(a, b), a < b.

1 x−a
PDF: f (x) = on (a, b); 0 else. CDF: F (x) = for x ∈ [a, b].
b−a b−a
a+b (b − a) 2 tb
e −e ta
Mean: . Var: . MGF: MX (t) = (t ̸= 0).
2 12 (b − a)t
Notes: Basis of inverse transform sampling.

Exponential(λ), λ > 0.
PDF: f (x) = λe−λx for x ≥ [Link]: F (x) = 1 − e−λx .
λ
Mean: 1/λ. Var: 1/λ2 . MGF: MX (t) = , t < λ.
λ−t
Notes: Memoryless. min of independent exponentials is exponential; sums ⇒ Gamma.

Gamma(α, λ) (shape α, rate λ), α, λ > 0.

λα α−1 −λx
PDF: f (x) = x e (x > 0). CDF: Regularized incomplete gamma.
Γ(α)
Mean: α/λ. Var: α/λ2 . MGF: MX (t) = (1 − λt )−α , t < λ.
Notes: Sum of i.i.d. Exponential(λ) gives Gamma(α = n, λ) (Erlang). χ2ν = Gamma(ν/2, 1/2).

Beta Type I (standard Beta) Beta(α, β) on (0, 1), α, β > 0.

xα−1 (1 − x)β−1
PDF: f (x) = , 0 < x < 1. CDF: Regularized incomplete beta.
B(α, β)
α αβ
Mean: . Var: 2
.
α+β (α + β) (α + β + 1)
MGF: No simple closed form. Notes: Conjugate prior for Bernoulli/Binomial p (Beta–Binomial).

Beta Type II (Beta Prime) Beta′ (α, β) on (0, ∞), α, β > 0.

xα−1
PDF: f (x) = , x > 0.
(1 + x)α+β B(α, β)
α α(α + β − 1)
Mean: for β > 1. Var: for β > 2.
β−1 (β − 2)(β − 1)2
MGF: No simple closed form.
U
Notes: If U ∼ Beta(α, β) then ∼ Beta′ (α, β). Arises as ratio of independent Gammas.
1−U

Normal N (µ, σ 2 ), σ > 0.

1  (x − µ)2 
. CDF: Φ x−µ


PDF: f (x) = √ exp − 2 σ (no simple closed form).
2πσ 2σ
Mean/Var: µ, σ 2 . MGF: MX (t) = exp µt + 21 σ 2 t2 .

Notes: Sums of independent Normals are Normal. CLT: sums of many i.i.d. ⇒ approx Normal.

Standard Normal N (0, 1).

1 2
PDF: ϕ(z) = √ e−z /2 . CDF: Φ(z). MGF: exp(t2 /2).
2π
X −µ
Notes: Standardization: if X ∼ N (µ, σ 2 ), then Z = ∼ N (0, 1).
σ

Cauchy(x0 , γ), γ > 0.

1 1 1 x−x0
+ 21 .


PDF: f (x) =
. CDF: F (x) = arctan γ
πγ 1 + x−x0 2 π
γ
Mean/Var: Undefined. MGF: Does not exist; CF φ(t) = eix0 t−γ|t| .
Notes: Heavy-tailed; sample mean does not converge to a finite value.

2
 Chi-square(ν), ν > 0.
1
PDF: f (x) = ν/2 xν/2−1 e−x/2 , x > 0. CDF: Regularized gamma.
2 Γ(ν/2)
Mean: ν. Var: 2ν. MGF: (1 − 2t)−ν/2 , t < 1/2.
iid P 2
Notes: If Zi ∼ N (0, 1), then Zi ∼ χ2ν .

Student’s t(ν), ν > 0.

Γ ν+1  x2 −(ν+1)/2
PDF: f (x) = √ 2 ν
1 + .
νπ Γ 2 ν
ν
Mean: 0 for ν > 1 (undefined for ν ≤ 1). Var: for ν > 2 (infinite for 1 < ν ≤ 2, undefined for
ν−2
ν ≤ 1).
MGF: Does not exist. CF exists.
Z
Notes: If Z ∼ N (0, 1) and V ∼ χ2ν independent, then T = p ∼ tν .
V /ν

F(d1 , d2 ), d1 , d2 > 0.
1  d d1 /2
1 xd1 /2−1
PDF: f (x) = , x > 0.
B( d21 , d22 ) d2 (1 + dd12 x)(d1 +d2 )/2
d2 2d22 (d1 + d2 − 2)
Mean: for d2 > 2. Var: for d2 > 4.
d2 − 2 d1 (d2 − 2)2 (d2 − 4)
MGF: No simple closed form.
U/d1 d2 2
Notes: If U ∼ χ2d1 , V ∼ χ2d2 indep., then ∼ F (d1 , d2 ). Also F (d1 , d2 ) = t with ν = d2 when
V /d2 d1 ν
d1 = 1.

Key Interrelations (Quick Hits)

• Bernoulli sum ⇒ Binomial. Poisson limit: Bin(n, p) → Pois(λ) if np → λ.

• Poisson thinning/superposition: sums stay Poisson; independent thinning with prob q gives Pois(qλ).

• Geometric is memoryless (discrete); Exponential is memoryless (continuous).

• Sum of exponentials ⇒ Gamma; Erlang when shape integer.

• Normal closed under linear combinations; CLT: standardized sums → Normal.

• χ2 , t, F from Normal via quadratic/ratio forms.

• Conjugacy: Beta–Binomial, Gamma–Poisson, Normal–Normal (known variance).

• With vs without replacement: Binomial (with replacement) ≈ Hypergeometric (without) when N

large.
U
• Beta/Beta’: If U ∼ Beta(α, β) then ∼ Beta′ (α, β).
1−U

3
One-Page Recap (Pin This)

Shapes & Supports • No simple/doesn’t exist: Hypergeometric (no sim-

ple), Beta/Beta’ (no simple), Cauchy (none), t
• Discrete: Uniform {a, . . . , b}; Bernoulli {0, 1}; Bi-
(none), F (no simple).
nomial 0:n; Poisson N0 ; Geometric 1:∞; Hypergeo
bounded; NegBin r:∞.
• Continuous: Uniform (a, b); Exponential (0, ∞); Signature Properties
Gamma (0, ∞); Beta (0, 1); Beta’ (0, ∞); Normal
• Memoryless: Geom/Exp.
R; Cauchy R; χ2 , (0, ∞); t, R; F, (0, ∞).
• Closure: Poisson sums; Normal linear combos.
Means & Variances (fast)
• Bernoulli (p, p(1 − p)); Binomial (np, np(1 − p)); • Limits: Binomial → Poisson; standardized sums
Poisson (λ, λ). → Normal (CLT).
• Geometric (1/p, (1 − p)/p2 ); NegBin (r/p, r(1 −
• Ratios/Quadratics: χ2 from Normal squares; t =
p)/p2 ).
√Z ; F = U/d 1
V /d2 .
V /ν
• Hypergeo n N , n N (1 − N ) N −1 .
K K K N −n

• Uniform cont. a+b (b−a)2
• Conjugacy: Beta–Binomial; Gamma–Poisson;

2 , 12 .
Normal–Normal.
• Exponential 1/λ, 1/λ2 ; Gamma α/λ, α/λ2 .

• Beta α+β α
, (α+β)2αβ


(α+β+1)
. Parameterizations (watch-outs)
• Normal (µ, σ 2 ); χ2 (ν, 2ν); t (mean 0 if ν > 1, • Geometric/NegBin: trials vs failures forms.
var ν/(ν − 2) if ν > 2); F (mean d2 /(d2 − 2) if
d2 > 2).
• Gamma: rate λ vs scale θ = 1/λ.

MGFs (existence quick check)

• Beta’: on (0, ∞), heavy right tail for small β.
• Exists/simple: Bernoulli, Binomial, Poisson, Geo-
metric/NegBin (domain t < − ln(1 − p)), Uniform, • Cauchy: no mean/variance; use median and quan-
Exponential, Gamma, Normal, χ2 . tiles.

Tip: memorize the “generator” relations (Poisson ↔ Gamma, Binomial ↔ Beta, Normal family) and the few MGFs
you actually use often; the rest can be derived on the fly.