Namespaces
Variants
Actions

Difference between revisions of "Steepest descent, method of"

From Encyclopedia of Mathematics
Jump to: navigation, search
(Importing text file)
 
m (tex/msc/mr/zbl)
 
Line 1: Line 1:
A special instance of the method of descent (cf. [[Descent, method of|Descent, method of]]), when the direction <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s0875701.png" /> of descent is chosen as the direction opposite to <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s0875702.png" />. The formulas of the method of steepest descent are
+
{{MSC|65}}
 +
{{TEX|done}}
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s0875703.png" /></td> </tr></table>
 
  
where the parameters <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s0875704.png" /> are determined by the condition that the function <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s0875705.png" /> has maximum decrease at each step. If <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s0875706.png" /> is twice continuously-differentiable and its matrix of second derivatives <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s0875707.png" /> satisfies the inequality
+
The method of steepest decent is
 +
a special instance of the method of descent (cf.
 +
[[Descent, method of|Descent, method of]]), when the direction $g^k$ of descent is chosen as the direction opposite to $\mathrm{grad} f(x^k)$. The formulas of the method of steepest descent are
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s0875708.png" /></td> </tr></table>
+
$$x^{k+1} = x^k - \def\a{\alpha} \a_k f'(x^k),\quad k = 0,1,\dots ,$$
 +
where the parameters $\{\a_k\}$ are determined by the condition that the function $f$ has maximum decrease at each step. If $f$ is twice continuously-differentiable and its matrix of second derivatives $f''$ satisfies the inequality
  
for any <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s0875709.png" />, with constants <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757010.png" />, then (see [[#References|[2]]], [[#References|[4]]]) the sequence <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757011.png" /> converges to a solution <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757012.png" /> of the problem of minimizing <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757013.png" />, the convergence rate being that of a geometric progression with quotient <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757014.png" />.
+
$$m\|y\|^2 \le (f''(x)y,y)\le M \|y\|^2$$
 +
for any $x,y$, with constants $M\ge m > 0$, then (see
 +
{{Cite|KaAk}},
 +
{{Cite|PsDa}}) the sequence $\{x^k\}$ converges to a solution $x^*$ of the problem of minimizing $f$, the convergence rate being that of a geometric progression with quotient $q<1$.
  
The method of steepest descent has been widely applied to the solution of systems of linear algebraic equations <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757015.png" /> with a Hermitian or positive-definite matrix <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757016.png" />. In the real symmetric case, the problem of solving this system is equivalent to finding a vector <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757017.png" /> in an <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757018.png" />-dimensional vector space minimizing the functional
+
The method of steepest descent has been widely applied to the solution of systems of linear algebraic equations $Ax = f$ with a Hermitian or positive-definite matrix $A$. In the real symmetric case, the problem of solving this system is equivalent to finding a vector $x^*$ in an $n$-dimensional vector space minimizing the functional
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757019.png" /></td> <td valign="top" style="width:5%;text-align:right;">(*)</td></tr></table>
 
  
 +
$$F(X)=\frac{1}{2} (Ax,x) -(x,f).\tag{$*$}$$
 
Applied to (*), the method of steepest descent takes the form
 
Applied to (*), the method of steepest descent takes the form
  
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757020.png" /></td> </tr></table>
+
$$x^{k+1} = x^k - \a_k(Ax^k -f), \quad k = 0,1,\dots ,$$
 
+
where the value of $\a_k$ is determined by minimization of the functional (*), according to the formula
where the value of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757021.png" /> is determined by minimization of the functional (*), according to the formula
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757022.png" /></td> </tr></table>
 
 
 
If the spectrum of <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757023.png" /> lies in an interval <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757024.png" />, <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757025.png" />, on the real axis, then the sequence <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757026.png" /> converges to the solution <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757027.png" /> at the rate of a geometric progression with quotient <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757028.png" />.
 
 
 
The method of steepest descent can be applied to solve an operator equation <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757029.png" /> with a self-adjoint positive-definite bounded operator <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757030.png" />. If <img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757031.png" /> does not satisfy these conditions, the problem can be symmetrized, reduced to the problem
 
 
 
<table class="eq" style="width:100%;"> <tr><td valign="top" style="width:94%;text-align:center;"><img align="absmiddle" border="0" src="https://www.encyclopediaofmath.org/legacyimages/s/s087/s087570/s08757032.png" /></td> </tr></table>
 
 
 
and then one can apply the method of steepest descent (see also [[Minimal discrepancy method|Minimal discrepancy method]]).
 
 
 
====References====
 
<table><TR><TD valign="top">[1]</TD> <TD valign="top">  L.V. Kantorovich,  "On the method of steepest descent"  ''Dokl. Akad. Nauk SSSR'' , '''56''' :  3  (1947)  pp. 233–236  (In Russian)</TD></TR><TR><TD valign="top">[2]</TD> <TD valign="top">  L.V. Kantorovich,  G.P. Akilov,  "Functional analysis" , Pergamon  (1982)  (Translated from Russian)</TD></TR><TR><TD valign="top">[3]</TD> <TD valign="top">  D.K. Faddeev,  V.N. Faddeeva,  "Computational methods of linear algebra" , Freeman  (1963)  (Translated from Russian)</TD></TR><TR><TD valign="top">[4]</TD> <TD valign="top">  B.N. Pshenichnyi,  Yu.M. Danilin,  "Numerical methods in extremal problems" , MIR  (1978)  (Translated from Russian)</TD></TR><TR><TD valign="top">[5]</TD> <TD valign="top">  N.S. Bakhvalov,  "Numerical methods: analysis, algebra, ordinary differential equations" , MIR  (1977)  (Translated from Russian)</TD></TR></table>
 
 
 
  
 +
$$\a_k = \frac{(\xi^k,\xi^k)}{(A\xi^k,\xi^k)},\quad \xi^k = Ax^k - f.$$
 +
If the spectrum of $A$ lies in an interval $[m,M]$, $M\ge m > 0$, on the real axis, then the sequence $\{x^k\}$ converges to the solution $x^*$ at the rate of a geometric progression with quotient $q=(M-m)/(M+m)$.
  
====Comments====
+
The method of steepest descent can be applied to solve an operator equation $Au=f$ with a self-adjoint positive-definite bounded operator $A$. If $A$ does not satisfy these conditions, the problem can be symmetrized, reduced to the problem
  
 +
$$A^* Au = A^*f,$$
 +
and then one can apply the method of steepest descent (see also
 +
[[Minimal discrepancy method|Minimal discrepancy method]]).
  
 
====References====
 
====References====
<table><TR><TD valign="top">[a1]</TD> <TD valign="top"> G.H. Golub,  C.F. van Loan,  "Matrix computations" , Johns Hopkins Univ. Press  (1989)</TD></TR></table>
+
{|
 +
|-
 +
|valign="top"|{{Ref|Ba}}||valign="top"|  N.S. Bakhvalov,  "Numerical methods: analysis, algebra, ordinary differential equations", MIR  (1977)  (Translated from Russian)  {{MR|0362811}}  {{ZBL|0524.65001}}
 +
|-
 +
|valign="top"|{{Ref|FaFa}}||valign="top"|  D.K. Faddeev,  V.N. Faddeeva,  "Computational methods of linear algebra", Freeman  (1963)  (Translated from Russian)  {{MR|0161454}} {{MR|0158519}}  {{ZBL|0451.65015}}
 +
|-
 +
|valign="top"|{{Ref|GoLo}}||valign="top"| G.H. Golub,  C.F. van Loan,  "Matrix computations", Johns Hopkins Univ. Press  (1989) {{MR|1002570}}  {{ZBL|0733.65016}}
 +
|-
 +
|valign="top"|{{Ref|Ka}}||valign="top"|  L.V. Kantorovich,  "On the method of steepest descent"  ''Dokl. Akad. Nauk SSSR'', '''56''' :  3  (1947)  pp. 233–236  (In Russian) 
 +
|-
 +
|valign="top"|{{Ref|KaAk}}||valign="top"|  L.V. Kantorovich,  G.P. Akilov,  "Functional analysis", Pergamon  (1982)  (Translated from Russian)  {{MR|0664597}}  {{ZBL|0484.46003}}
 +
|-
 +
|valign="top"|{{Ref|PsDa}}||valign="top"|  B.N. Pshenichnyi,  Yu.M. Danilin,  "Numerical methods in extremal problems", MIR  (1978)  (Translated from Russian)  {{MR|0502886}} 
 +
|-
 +
|}

Latest revision as of 19:10, 6 April 2012

2020 Mathematics Subject Classification: Primary: 65-XX [MSN][ZBL]


The method of steepest decent is a special instance of the method of descent (cf. Descent, method of), when the direction $g^k$ of descent is chosen as the direction opposite to $\mathrm{grad} f(x^k)$. The formulas of the method of steepest descent are

$$x^{k+1} = x^k - \def\a{\alpha} \a_k f'(x^k),\quad k = 0,1,\dots ,$$ where the parameters $\{\a_k\}$ are determined by the condition that the function $f$ has maximum decrease at each step. If $f$ is twice continuously-differentiable and its matrix of second derivatives $f''$ satisfies the inequality

$$m\|y\|^2 \le (f''(x)y,y)\le M \|y\|^2$$ for any $x,y$, with constants $M\ge m > 0$, then (see [KaAk], [PsDa]) the sequence $\{x^k\}$ converges to a solution $x^*$ of the problem of minimizing $f$, the convergence rate being that of a geometric progression with quotient $q<1$.

The method of steepest descent has been widely applied to the solution of systems of linear algebraic equations $Ax = f$ with a Hermitian or positive-definite matrix $A$. In the real symmetric case, the problem of solving this system is equivalent to finding a vector $x^*$ in an $n$-dimensional vector space minimizing the functional

$$F(X)=\frac{1}{2} (Ax,x) -(x,f).\tag{$*$}$$ Applied to (*), the method of steepest descent takes the form

$$x^{k+1} = x^k - \a_k(Ax^k -f), \quad k = 0,1,\dots ,$$ where the value of $\a_k$ is determined by minimization of the functional (*), according to the formula

$$\a_k = \frac{(\xi^k,\xi^k)}{(A\xi^k,\xi^k)},\quad \xi^k = Ax^k - f.$$ If the spectrum of $A$ lies in an interval $[m,M]$, $M\ge m > 0$, on the real axis, then the sequence $\{x^k\}$ converges to the solution $x^*$ at the rate of a geometric progression with quotient $q=(M-m)/(M+m)$.

The method of steepest descent can be applied to solve an operator equation $Au=f$ with a self-adjoint positive-definite bounded operator $A$. If $A$ does not satisfy these conditions, the problem can be symmetrized, reduced to the problem

$$A^* Au = A^*f,$$ and then one can apply the method of steepest descent (see also Minimal discrepancy method).

References

[Ba] N.S. Bakhvalov, "Numerical methods: analysis, algebra, ordinary differential equations", MIR (1977) (Translated from Russian) MR0362811 Zbl 0524.65001
[FaFa] D.K. Faddeev, V.N. Faddeeva, "Computational methods of linear algebra", Freeman (1963) (Translated from Russian) MR0161454 MR0158519 Zbl 0451.65015
[GoLo] G.H. Golub, C.F. van Loan, "Matrix computations", Johns Hopkins Univ. Press (1989) MR1002570 Zbl 0733.65016
[Ka] L.V. Kantorovich, "On the method of steepest descent" Dokl. Akad. Nauk SSSR, 56 : 3 (1947) pp. 233–236 (In Russian)
[KaAk] L.V. Kantorovich, G.P. Akilov, "Functional analysis", Pergamon (1982) (Translated from Russian) MR0664597 Zbl 0484.46003
[PsDa] B.N. Pshenichnyi, Yu.M. Danilin, "Numerical methods in extremal problems", MIR (1978) (Translated from Russian) MR0502886
How to Cite This Entry:
Steepest descent, method of. Encyclopedia of Mathematics. URL: http://encyclopediaofmath.org/index.php?title=Steepest_descent,_method_of&oldid=13529
This article was adapted from an original article by Yu.A. Kuznetsov (originator), which appeared in Encyclopedia of Mathematics - ISBN 1402006098. See original article