TOPICS
Search

Square Root Method


The square root method is an algorithm which solves the matrix equation

 Au=g
(1)

for u, with A a p×p symmetric matrix and g a given vector. Convert A to a triangular matrix such that

 T^(T)T=A,
(2)

where T^(T) is the transpose. Then

T^(T)k=g
(3)
Tu=k,
(4)

so

 T=[s_(11) s_(12) ... ...; 0 s_(22) ... ...; | | ... |; 0 0 ... s_(pp)],
(5)

giving the equations

s_(11)^2=a_(11)
(6)
s_(11)s_(12)=a_(12)
(7)
s_(12)^2+s_(22)^2=a_(22)
(8)
s_(1j)^2+s_(2j)^2+...+s_(jj)^2=a_(jj)
(9)
s_(1j)+s_(2j)s_(2k)+...+s_(jj)s_(jk)=a_(jk).
(10)

These give

s_(11)=sqrt(a_(11))
(11)
s_(12)=(a_(12))/(s_(11))
(12)
s_(22)=sqrt(a_(22)-s_(12)^2)
(13)
s_(jj)=sqrt(a_(jj)-s_(ij)^2-s_(2j)^2-...-s_(j-1,j)^2)
(14)
s_(jk)=(a_(jk)-s_(1j)s_(1k)-s_(2j)s_(2k)-...-s_(j-1,j)s_(j-1,k))/(s_(jj)),
(15)

giving T from A. Now solve for k in terms of the s_(ij)s and g,

s_(11)k_1=g_1
(16)
s_(12)k_1+s_(22)k_2=g_2
(17)
s_(1j)k_1+s_(2j)k_2+...+s_(jj)k_j=g_j,
(18)

which gives

k_1=(g_1)/(s_(11))
(19)
k_2=(g_2-s_(12)k_1)/(s_(22))
(20)
k_j=(g_j-s_(1j)k_1-s_(2j)k_2-...-s_(j-1,j)k_(j-1))/(s_(jj)).
(21)

Finally, find u from the s_(ij)s and k,

s_(11)u_1+s_(12)u_2...+s_(1p)u_p=k_1
(22)
s_(22)u_2+...+s_(2p)u_p=k_2
(23)
s_(pp)u_p=k_p,
(24)

giving the desired solution,

u_p=(k_p)/(s_(pp))
(25)
u_(p-1)=(k_(p-1)-s_(p-1,p)u_p)/(s_(p-1,p-1))
(26)
u_j=(k_j-s_(j,j+1)u_(j+1)-s_(j,j+2)u_(j+2)-...-s_(jp)u_p)/(s_(jj)).
(27)

See also

LU Decomposition

Explore with Wolfram|Alpha

References

Kenney, J. F. and Keeping, E. S. Mathematics of Statistics, Pt. 2, 2nd ed. Princeton, NJ: Van Nostrand, pp. 298-300, 1951.

Referenced on Wolfram|Alpha

Square Root Method

Cite this as:

Weisstein, Eric W. "Square Root Method." From MathWorld--A Wolfram Web Resource. https://mathworld.wolfram.com/SquareRootMethod.html

Subject classifications