Phân phối chuẩn nhiều chiều

Phân phối chuẩn nhiều chiều
Hàm mật độ xác suất
Hàm phân phối tích lũy
Tham số μ = [ μ 1 , , μ N ] T {\displaystyle \mu =[\mu _{1},\dots ,\mu _{N}]^{T}} vị trí (véc tơ giá trị thực)
Σ {\displaystyle \Sigma } ma trận hiệp phương sai (ma trận xác định dương giá trị thực kích thước N × N {\displaystyle N\times N} )
Giá x R N {\displaystyle x\in \mathbb {R} ^{N}\!}
Hàm mật độ xác suất f X ( x 1 , , x N ) = 1 ( 2 π ) N / 2 | Σ | 1 / 2 {\displaystyle f_{X}(x_{1},\dots ,x_{N})={\frac {1}{(2\pi )^{N/2}\left|\Sigma \right|^{1/2}}}}
exp ( 1 2 ( x μ ) Σ 1 ( x μ ) ) {\displaystyle \exp \left(-{\frac {1}{2}}(x-\mu )^{\top }\Sigma ^{-1}(x-\mu )\right)}
Hàm phân phối tích lũy
Giá trị kỳ vọng μ {\displaystyle \mu }
Trung vị μ {\displaystyle \mu }
Yếu vị μ {\displaystyle \mu }
Phương sai Σ {\displaystyle \Sigma } (ma trận hiệp phương sai)
Độ xiên 0
Độ nhọn 0
Entropy ln ( ( 2 π e ) N | Σ | ) {\displaystyle \ln \left({\sqrt {(2\,\pi \,e)^{N}\left|\Sigma \right|}}\right)\!}
Hàm sinh mô men M X ( t ) = exp ( μ t + 1 2 t Σ t ) {\displaystyle M_{X}(t)=\exp \left(\mu ^{\top }t+{\frac {1}{2}}t^{\top }\Sigma t\right)}
Hàm đặc trưng ϕ X ( t ; μ , Σ ) = exp ( i μ t 1 2 t Σ t ) {\displaystyle \phi _{X}(t;\mu ,\Sigma )=\exp \left(i\mu ^{\top }t-{\frac {1}{2}}t^{\top }\Sigma t\right)}


Trong lý thuyết xác suấtthống kê, phân phối chuẩn nhiều chiều, đôi khi được gọi là phân phối Gauss nhiều chiều, là tổng quát hóa của phân phối chuẩn một chiều (còn gọi là phân phối Gauss) cho không gian nhiều chiều hơn. Phân phối này còn có quan hệ gần gũi với phân phối chuẩn ma trận.

Trường hợp tổng quát

Một véc tơ ngẫu nhiên   X = [ X 1 , , X N ] T {\displaystyle \ X=[X_{1},\dots ,X_{N}]^{T}} tuân theo một phân phối chuẩn nhiều chiều nếu nó thỏa mãn các điều kiện tương đương nhau sau đây:

  • mọi tổ hợp tuyến tính   Y = a 1 X 1 + + a N X N {\displaystyle \ Y=a_{1}X_{1}+\cdots +a_{N}X_{N}} đều tuân theo phân phối chuẩn
  • tồn tại một véc tơ ngẫu nhiên   Z = [ Z 1 , , Z M ] T {\displaystyle \ Z=[Z_{1},\dots ,Z_{M}]^{T}} , trong đó các thành phần của nó là các biến ngẫu nhiên độc lập tuân theo phân phối chuẩn chuẩn hóa, một véc tơ   μ = [ μ 1 , , μ N ] T {\displaystyle \ \mu =[\mu _{1},\dots ,\mu _{N}]^{T}} và một ma trận   A {\displaystyle \ A} kích thước N × M {\displaystyle N\times M} sao cho   X = A Z + μ {\displaystyle \ X=AZ+\mu } .
  • tồn tại một véc tơ μ {\displaystyle \mu } và một ma trận   Σ {\displaystyle \ \Sigma } đối xứng, nửa xác định dương sao cho hàm đặc trưng của X
ϕ X ( u ; μ , Σ ) = exp ( i μ u 1 2 u Σ u ) . {\displaystyle \phi _{X}\left(u;\mu ,\Sigma \right)=\exp \left(i\mu ^{\top }u-{\frac {1}{2}}u^{\top }\Sigma u\right).}

Nếu   Σ {\displaystyle \ \Sigma } là ma trận không suy biến, thì phân phối này có thể được mô tả bởi hàm mật độ xác suất sau:

f X ( x 1 , , x N ) = 1 ( 2 π ) N / 2 | Σ | 1 / 2 exp ( 1 2 ( x μ ) Σ 1 ( x μ ) ) {\displaystyle f_{X}(x_{1},\dots ,x_{N})={\frac {1}{(2\pi )^{N/2}|\Sigma |^{1/2}}}\exp \left(-{\frac {1}{2}}(x-\mu )^{\top }\Sigma ^{-1}(x-\mu )\right)}

trong đó   | Σ | {\displaystyle \ \left|\Sigma \right|} định thức của   Σ {\displaystyle \ \Sigma } . Lưu ý rằng phương trình trên suy biến về phương trình của phân phối chuẩn một chiều nếu   Σ {\displaystyle \ \Sigma } là một giá trị vô hướng (nghĩa là một ma trận 1x1).

Véc tơ μ trong các điều kiện trên là giá trị kỳ vọng của X và ma trận   Σ = A A T {\displaystyle \ \Sigma =AA^{T}} là ma trận hiệp phương sai của thành phần Xi.

Cần lưu ý rằng ma trận hiệp phương sai có thể suy biến (và khi đó không được mô tả bởi các công thức sử dụng   Σ 1 {\displaystyle \ \Sigma ^{-1}} ở trên).

Trường hợp này thường xảy ra trong thống kê; ví dụ, trong phân phối của véc tơ dư trong các bài toán hồi quy tuyến tính thông thường. Cũng lưu ý rằng các Xi nói chung là không độc lập; chúng có thể được xem là kết quả của việc áp dụng biến đổi tuyến tính A cho tập hợp Z gồm các biến ngẫu nhiên Gauss độc lập.

Việc phân phối của một véc tơ ngẫu nhiên X là một phân phối chuẩn nhiều chiều được ký hiệu bởi công thức sau:

X   N ( μ , Σ ) , {\displaystyle X\ \sim {\mathcal {N}}(\mu ,\Sigma ),}

hoặc viết tường minh rằng X biến trong không gian N-chiều,

X   N N ( μ , Σ ) . {\displaystyle X\ \sim {\mathcal {N}}_{N}(\mu ,\Sigma ).}

Hàm phân phối tích lũy

Hàm phân phối tích lũy (cdf) F ( x ) {\displaystyle F(x)} được định nghĩa là xác suất mà mỗi giá trị trong một véc tơ ngẫu nhiên X {\displaystyle X} đều nhỏ hơn hay bằng giá trị tương ứng trong véc tơ x {\displaystyle x} . Tuy không có dạng đóng cho F ( x ) {\displaystyle F(x)} , có một số thuật toán ước tính giá trị của nó. Ví dụ, xem MVNDST tại [1] (dùng FORTRAN) hay [2] Lưu trữ 2008-05-13 tại Wayback Machine (dùng MATLAB).

Một phản ví dụ

Điều kiện rằng hai biến ngẫu nhiên XY đều có phân phối chuẩn không kéo theo việc cặp (XY) có một phân phối chuẩn có điều kiện phụ thuộc (joint normal distribution). Một ví dụ đơn giản là: Y = X nếu |X| > 1 và Y = −X nếu |X| < 1. Điều này cũng đúng cho số biến ngẫu nhiên nhiều hơn 2.

Độc lập và phân phối chuẩn

Nếu XY có phân phối chuẩn và độc lập thống kê, thì chúng có một phân phối chuẩn có điều kiện phụ thuộc, nghĩa là cặp (XY) phải có phân phối chuẩn 2 chiều. Tuy nhiên, một cặp biến ngẫu nhiên có phân phối chuẩn có điều kiện phụ thuộc không nhất thiết độc lập lẫn nhau.

Trường hợp 2 chiều

Trong trường hợp 2 chiều không suy biến, hàm mật độ xác suất (với kì vọng (0,0)) là

f ( x , y ) = 1 2 π σ x σ y 1 ρ 2 exp ( 1 2 ( 1 ρ 2 ) ( x 2 σ x 2 + y 2 σ y 2 2 ρ x y ( σ x σ y ) ) ) {\displaystyle f(x,y)={\frac {1}{2\pi \sigma _{x}\sigma _{y}{\sqrt {1-\rho ^{2}}}}}\exp \left(-{\frac {1}{2(1-\rho ^{2})}}\left({\frac {x^{2}}{\sigma _{x}^{2}}}+{\frac {y^{2}}{\sigma _{y}^{2}}}-{\frac {2\rho xy}{(\sigma _{x}\sigma _{y})}}\right)\right)}

trong đó ρ {\displaystyle \rho } là tương quan giữa X {\displaystyle X} Y {\displaystyle Y} . Khi đó,

Σ = [ σ x 2 ρ σ x σ y ρ σ x σ y σ y 2 ] {\displaystyle \Sigma ={\begin{bmatrix}\sigma _{x}^{2}&\rho \sigma _{x}\sigma _{y}\\\rho \sigma _{x}\sigma _{y}&\sigma _{y}^{2}\end{bmatrix}}} .

Biến đổi afin

Nếu Y = c + B X {\displaystyle Y=c+BX\,} là một biến đổi afin của X   N ( μ , Σ ) , {\displaystyle X\ \sim {\mathcal {N}}(\mu ,\Sigma ),} trong đó c {\displaystyle c\,} là một véc tơ M × 1 {\displaystyle M\times 1} gồm các hằng số và B {\displaystyle B\,} là ma trận M × N {\displaystyle M\times N} , thì Y {\displaystyle Y\,} có phân phối chuẩn nhiều chiều với giá trị kỳ vọng c + B μ {\displaystyle c+B\mu \,} và phương sai B Σ B T {\displaystyle B\Sigma B^{T}\,} nghĩa là, Y N ( c + B μ , B Σ B T ) {\displaystyle Y\sim {\mathcal {N}}\left(c+B\mu ,B\Sigma B^{T}\right)} . Đặc biệt, tập con bất kỳ của X i {\displaystyle X_{i}\,} đều có một phân phối biên duyên là phân phối chuẩn nhiều chiều. Để minh họa, ta xét ví dụ sau: để tách tập con ( X 1 , X 2 , X 4 ) T {\displaystyle (X_{1},X_{2},X_{4})^{T}\,} , sử dụng

B = [ 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 ] {\displaystyle B={\begin{bmatrix}1&0&0&0&0&\ldots &0\\0&1&0&0&0&\ldots &0\\0&0&0&1&0&\ldots &0\end{bmatrix}}}

ma trận này trích lấy các phần tử mong muốn.

Một hệ quả khác là phân phối của Z = b X {\displaystyle Z=b\cdot X} , trong đó b {\displaystyle b} là một véc tơ có cùng số chiều với X {\displaystyle X} và dấu chấm ký hiệu phép nhân véc tơ, là phân phối chuẩn một chiều với Z N ( b μ , b T Σ b ) {\displaystyle Z\sim {\mathcal {N}}\left(b\cdot \mu ,b^{T}\Sigma b\right)} . Kết quả đó thu được bằng cách sử dụng

B = [ b 1 b 2 b n 0 0 0 0 0 0 ] {\displaystyle B={\begin{bmatrix}b_{1}&b_{2}&\ldots &b_{n}\\0&0&\ldots &0\\&&\vdots &\\0&0&\ldots &0\\\end{bmatrix}}}

và chỉ xét thành phần đầu tiên của tích (hàng đầu của B {\displaystyle B} là véc tơ b {\displaystyle b} ). Để ý tính chất xác định dương của Σ {\displaystyle \Sigma } hàm ý rằng phương sai của tích vô hướng phải là số dương.

Tham khảo

Đọc thêm

  • Rencher, A.C. (1995). Methods of Multivariate Analysis. New York: Wiley.
  • Tong, Y. L. (1990). The multivariate normal distribution. Springer Series in Statistics. New York: Springer-Verlag. doi:10.1007/978-1-4613-9655-0. ISBN 978-1-4613-9657-4.
  • x
  • t
  • s
Phân phối xác suất (Danh sách phân phối xác suất)
Discrete univariate
with finite support
Đơn biến rời rạc
với hỗ trợ vô hạn
  • Beta negative binomial distribution
  • Borel distribution
  • Conway–Maxwell–Poisson distribution
  • Discrete phase-type distribution
  • Delaporte distribution
  • Extended negative binomial distribution
  • Flory–Schulz distribution
  • Gauss–Kuzmin distribution
  • Phân phối hình học
  • Logarithmic distribution
  • Negative binomial distribution
  • (a,b,0) class of distributions
  • Parabolic fractal distribution
  • Phân phối Poisson
  • Skellam distribution
  • Yule–Simon distribution
  • Zeta distribution
Continuous univariate
supported on a bounded interval
  • Arcsine distribution
  • ARGUS distribution
  • Balding–Nichols model
  • Bates distribution
  • Beta distribution
  • Beta rectangular distribution
  • Continuous Bernoulli distribution
  • Irwin–Hall distribution
  • Kumaraswamy distribution
  • Logit-normal distribution
  • Noncentral beta distribution
  • PERT distribution
  • Raised cosine distribution
  • Reciprocal distribution
  • Triangular distribution
  • U-quadratic distribution
  • Phân phối đều liên tục
  • Wigner semicircle distribution
Continuous univariate
supported on a semi-infinite interval
  • Benini distribution
  • Benktander type I distribution
  • Benktander type II distribution
  • Beta prime distribution
  • Burr distribution
  • Chi-square distribution
  • Chi distribution
  • Dagum distribution
  • Davis distribution
  • Exponential-logarithmic distribution
  • Erlang distribution
  • Phân phối mũ
  • F-distribution
  • Folded normal distribution
  • Fréchet distribution
  • Gamma distribution
  • Gamma/Gompertz distribution
  • Generalized gamma distribution
  • Generalized inverse Gaussian distribution
  • Gompertz distribution
  • Half-logistic distribution
  • Half-normal distribution
  • Hotelling's T-squared distribution
  • Hyper-Erlang distribution
  • Hyperexponential distribution
  • Hypoexponential distribution
  • Inverse-chi-squared distribution
    • Scaled inverse chi-squared distribution
  • Inverse Gaussian distribution
  • Inverse-gamma distribution
  • Kolmogorov–Smirnov test
  • Lévy distribution
  • Log-Cauchy distribution
  • Log-Laplace distribution
  • Log-logistic distribution
  • Log-normal distribution
  • Lomax distribution
  • Matrix-exponential distribution
  • Maxwell–Boltzmann distribution
  • Maxwell–Jüttner distribution
  • Mittag-Leffler distribution
  • Nakagami distribution
  • Noncentral chi-squared distribution
  • Noncentral F-distribution
  • Pareto distribution
  • Phase-type distribution
  • Poly-Weibull distribution
  • Rayleigh distribution
  • Relativistic Breit–Wigner distribution
  • Rice distribution
  • Shifted Gompertz distribution
  • Truncated normal distribution
  • Type-2 Gumbel distribution
  • Weibull distribution
    • Discrete Weibull distribution
  • Wilks's lambda distribution
Continuous univariate
supported on the whole real line
  • Cauchy distribution
  • Generalized normal distribution
  • Fisher's z-distribution
  • Gaussian q-distribution
  • Generalized normal distribution
  • Generalised hyperbolic distribution
  • Geometric stable distribution
  • Gumbel distribution
  • Holtsmark distribution
  • Hyperbolic secant distribution
  • Johnson's SU-distribution
  • Landau distribution
  • Laplace distribution
  • Asymmetric Laplace distribution
  • Logistic distribution
  • Noncentral t-distribution
  • Phân phối chuẩn
  • Normal-inverse Gaussian distribution
  • Skew normal distribution
  • Slash distribution
  • Stable distribution
  • Student's t-distribution
  • Type-1 Gumbel distribution
  • Tracy–Widom distribution
  • Variance-gamma distribution
  • Voigt profile
Continuous univariate
with support whose type varies
  • Generalized chi-squared distribution
  • Generalized extreme value distribution
  • Generalized Pareto distribution
  • Marchenko–Pastur distribution
  • Q-exponential distribution
  • Q-Gaussian distribution
  • Q-Weibull distribution
  • Shifted log-logistic distribution
  • Tukey lambda distribution
Mixed continuous-discrete univariate
  • Rectified Gaussian distribution
Phân phối xác suất đồng thời
Discrete
Ewens's sampling formula
Multinomial distribution
Dirichlet-multinomial distribution
Negative multinomial distribution
Continuous
Dirichlet distribution
Generalized Dirichlet distribution
Multivariate Laplace distribution
Phân phối chuẩn nhiều chiều
Multivariate stable distribution
Multivariate t-distribution
Normal-inverse-gamma distribution
Normal-gamma distribution
Random matrix
Inverse matrix gamma distribution
Inverse-Wishart distribution
Matrix normal distribution
Matrix t-distribution
Matrix gamma distribution
Normal-inverse-Wishart distribution
Normal-Wishart distribution
Wishart distribution
Directional statistics
Univariate (circular) Directional statistics
Circular uniform distribution
Von Mises distribution
Wrapped normal distribution
Wrapped Cauchy distribution
Wrapped exponential distribution
Wrapped asymmetric Laplace distribution
Wrapped Lévy distribution
Bivariate (spherical)
Kent distribution
Bivariate (toroidal)
Bivariate von Mises distribution
Multivariate
Von Mises–Fisher distribution
Bingham distribution
Degenerate distribution and Singular distribution
Degenerate
Hàm delta Dirac
Singular
Cantor distribution
Families
  • Circular distribution
  • Compound Poisson distribution
  • Elliptical distribution
  • Exponential family
  • Natural exponential family
  • Location–scale family
  • Maximum entropy probability distribution
  • Mixture distribution
  • Pearson distribution
  • Tweedie distribution
  • Wrapped distribution