<?xml-stylesheet type="text/xsl" href="../styles/pmathml.xsl"?>
<!-- saved from url=(0014)about:internet -->
<html xmlns="http://www.w3.org/1999/xhtml" xmlns:dsi="http://www.w3.org/1999/xlink" xmlns:m="http://www.w3.org/1998/Math/MathML" xml:space="preserve"><head><meta http-equiv="Content-Type" content="text/html; charset=US-ASCII"/><title>F01ABF : NAG Library, Mark 22</title><link rel="stylesheet" href="../styles/libdoc.css" type="text/css"/><script type="text/javascript">
   function showLevel(_levelId){
    var thisLevel = document.getElementById(_levelId);
    var thisplus = document.getElementById( _levelId.concat('plus'));
    var thisminus = document.getElementById( _levelId.concat('minus'));
    if(thisLevel.style.display != "block"){
     thisLevel.style.display = "block";
     thisplus.style.display = "none";
     thisminus.style.display = "inline";
     }
    else{
     thisLevel.style.display = "none";
     thisminus.style.display = "none";
     thisplus.style.display = "inline";
     }
    }
  </script></head><body><hr/><div><a class="rout" href="../../pdf/F01/f01abf.pdf">F01ABF (PDF version)</a></div><div><a class="chap" href="f01conts.xml">F01 Chapter Contents</a></div><div><a class="chapint" href="f01intro.xml">F01 Chapter Introduction</a></div>
<div><a class="htmltoc" href="../FRONTMATTER/manconts.xml">NAG Library Manual</a></div><hr/><h1 class="libdoc">NAG Library Routine Document<br/><br/>F01ABF</h1><div class="paramtext"><div class="header"><b>Note:</b>&#160; before using this routine, please read the Users' Note for your implementation to check the interpretation of <span class="bitalic">bold italicised</span> terms and other implementation-dependent details.</div></div> 
<div class="htmltoc">
<h2 class="htmltoc"><span class="htmltochead" onclick="showLevel('htmltoc');"><span class="htmltocplus" id="htmltocplus">+</span><span class="htmltocminus" id="htmltocminus">&#8722;</span></span>&#160;Contents</h2>
<div class="htmltocitem" id="htmltoc">
<div class="htmltoc">
<span class="htmltocplus">&#160;&#160;&#160;</span>
<a class="htmltoc" href="#purpose">1&#160;&#160;<b>Purpose</b></a>
</div><div class="htmltoc">
<span class="htmltocplus">&#160;&#160;&#160;</span>
<a class="htmltoc" href="#specification">2&#160;&#160;<b>Specification</b></a>
</div><div class="htmltoc">
<span class="htmltocplus">&#160;&#160;&#160;</span>
<a class="htmltoc" href="#description">3&#160;&#160;<b>Description</b></a>
</div><div class="htmltoc">
<span class="htmltocplus">&#160;&#160;&#160;</span>
<a class="htmltoc" href="#references">4&#160;&#160;<b>References</b></a>
</div><div class="htmltoc">
<span class="htmltocplus">&#160;&#160;&#160;</span>
<a class="htmltoc" href="#parameters">5&#160;&#160;<b>Parameters</b></a>
</div><div class="htmltoc">
<span class="htmltocplus">&#160;&#160;&#160;</span>
<a class="htmltoc" href="#errors">6&#160;&#160;<b>Error Indicators and Warnings</b></a>
</div><div class="htmltoc">
<span class="htmltocplus">&#160;&#160;&#160;</span>
<a class="htmltoc" href="#accuracy">7&#160;&#160;<b>Accuracy</b></a>
</div><div class="htmltoc">
<span class="htmltocplus">&#160;&#160;&#160;</span>
<a class="htmltoc" href="#fcomments">8&#160;&#160;<b>Further Comments</b></a>
</div><div class="htmltoc">
<span class="htmltoc" onclick="showLevel('tocexample');"><span class="htmltocplus" id="tocexampleplus">+</span><span class="htmltocminus" id="tocexampleminus">&#8722;</span></span>
<a class="htmltoc" href="#example">9&#160;&#160;<b>Example</b></a>
<div class="htmltocitem" id="tocexample">
<div class="htmltoc">
<span class="htmltocplus">&#160;&#160;&#160;</span>
<a class="htmltoc" href="#examtext">9.1&#160;&#160;<b>Program Text</b></a>
</div><div class="htmltoc">
<span class="htmltocplus">&#160;&#160;&#160;</span>
<a class="htmltoc" href="#examdata">9.2&#160;&#160;<b>Program Data</b></a>
</div><div class="htmltoc">
<span class="htmltocplus">&#160;&#160;&#160;</span>
<a class="htmltoc" href="#examresults">9.3&#160;&#160;<b>Program Results</b></a>
</div>
</div>
</div>
</div>
</div><h2 class="standard"><a class="sec" name="purpose" id="purpose"/>1&#160;&#160;Purpose</h2>
<div class="paramtext">F01ABF calculates the accurate inverse of a real symmetric positive-definite matrix, using a Cholesky factorization and iterative refinement.</div><h2 class="standard"><a class="sec" name="specification" id="specification"/>2&#160;&#160;Specification</h2>
<table class="fspec"><tr><td class="tdfspec1">SUBROUTINE&#160;F01ABF&#160;(</td><td class="tdfspec2"><a class="arg" href="#A">A</a>, <a class="arg" href="#LDA">LDA</a>, <a class="arg" href="#N">N</a>, <a class="arg" href="#B">B</a>, <a class="arg" href="#LDB">LDB</a>, <a class="arg" href="#Z">Z</a>, <a class="arg" href="#IFAIL">IFAIL</a>)</td></tr><tr><td class="tdfspec1">INTEGER</td><td class="tdfspec2">LDA, N, LDB, IFAIL</td></tr><tr><td class="tdfspec1"><b><i>double&#160;precision</i></b></td><td class="tdfspec2">A(LDA,N), B(LDB,N), Z(N)</td></tr></table><h2 class="standard"><a class="sec" name="description" id="description"/>3&#160;&#160;Description</h2>
<div class="paramtext">To compute the inverse <m:math><m:mi>X</m:mi></m:math>&#160;of a real symmetric positive-definite matrix <m:math><m:mi>A</m:mi></m:math>, F01ABF first computes a Cholesky factorization of <m:math><m:mi>A</m:mi></m:math>&#160;as <m:math><m:mi>A</m:mi><m:mo>=</m:mo><m:mi>L</m:mi><m:msup><m:mi>L</m:mi><m:mi mathvariant="normal">T</m:mi></m:msup></m:math>, where <m:math><m:mi>L</m:mi></m:math>&#160;is lower triangular.  An approximation to <m:math><m:mi>X</m:mi></m:math>&#160;is found by computing  <m:math><m:msup><m:mi>L</m:mi><m:mrow><m:mo>-</m:mo><m:mn>1</m:mn></m:mrow></m:msup></m:math>&#160;and then the product <m:math><m:msup><m:mi>L</m:mi><m:mrow><m:mo>-</m:mo><m:mi mathvariant="normal">T</m:mi></m:mrow></m:msup><m:msup><m:mi>L</m:mi><m:mrow><m:mo>-</m:mo><m:mn>1</m:mn></m:mrow></m:msup></m:math>.  The residual matrix <m:math><m:mi>R</m:mi><m:mo>=</m:mo><m:mi>I</m:mi><m:mo>-</m:mo><m:mi>A</m:mi><m:mi>X</m:mi></m:math>&#160;is calculated using <span class="bitalic">additional precision</span>,  and a correction <m:math><m:mi>D</m:mi></m:math>&#160;to <m:math><m:mi>X</m:mi></m:math>&#160;is found by solving  <m:math><m:mi>L</m:mi><m:msup><m:mi>L</m:mi><m:mi mathvariant="normal">T</m:mi></m:msup><m:mi>D</m:mi><m:mo>=</m:mo><m:mi>R</m:mi></m:math>.  <m:math><m:mi>X</m:mi></m:math>&#160;is replaced by  <m:math><m:mi>X</m:mi><m:mo>+</m:mo><m:mi>D</m:mi></m:math>, and this iterative refinement of the inverse is repeated until full machine accuracy has been obtained.</div><h2 class="standard"><a class="sec" name="references" id="references"/>4&#160;&#160;References</h2><div class="paramtext"><a name="ref103" id="ref103"/>Wilkinson J H and Reinsch C (1971)  <i>Handbook for Automatic Computation II, Linear Algebra</i> Springer&#8211;Verlag </div><h2 class="standard"><a class="sec" name="parameters" id="parameters"/>5&#160;&#160;Parameters</h2>
<dl><dt class="paramhead"><a name="A" id="A"/>1: &#160;&#160;&#8194; A(<a class="arg" href="#LDA">LDA</a>,<a class="arg" href="#N">N</a>) &#8211; <span class="bitalic">double precision</span> array<span class="pclass">Input/Output</span></dt><dd><div class="paramtext"><i>On entry</i>: the upper triangle of the <m:math><m:mi>n</m:mi></m:math>&#160;by <m:math><m:mi>n</m:mi></m:math>&#160;positive-definite symmetric matrix <m:math><m:mi>A</m:mi></m:math>. The elements of the array below the diagonal need not be set.</div>
<div class="paramtext"><i>On exit</i>: the lower triangle of the inverse matrix <m:math><m:mi>X</m:mi></m:math>&#160;is stored in the elements of the array below the diagonal, in rows <m:math><m:mn>2</m:mn></m:math>&#160;to <m:math><m:mi>n</m:mi><m:mo>+</m:mo><m:mn>1</m:mn></m:math>; <m:math><m:msub><m:mi>x</m:mi><m:mrow><m:mi>i</m:mi><m:mi>j</m:mi></m:mrow></m:msub></m:math>&#160;is stored in <m:math><m:mrow><m:maction actiontype="link" dsi:type="simple" dsi:href="#A"><m:mi mathcolor="#EE0000" mathvariant="bold">A</m:mi></m:maction><m:mfenced separators="," open="(" close=")"><m:mrow><m:mi>i</m:mi><m:mo>+</m:mo><m:mn>1</m:mn></m:mrow><m:mi>j</m:mi></m:mfenced></m:mrow></m:math>&#160;for <m:math><m:mi>i</m:mi><m:mo>&#8805;</m:mo><m:mi>j</m:mi></m:math>. The upper triangle of the original matrix is unchanged.</div></dd><dt class="paramhead"><a name="LDA" id="LDA"/>2: &#160;&#160;&#8194; LDA &#8211; INTEGER<span class="pclass">Input</span></dt><dd><div class="paramtext"><i>On entry</i>: the first dimension of the array <a class="arg" href="#A">A</a> as declared in the (sub)program from which F01ABF is called.</div><div class="paramtext"><i>Constraint</i>:
  <m:math><m:maction actiontype="link" dsi:type="simple" dsi:href="#LDA"><m:mi mathcolor="#EE0000" mathvariant="bold">LDA</m:mi></m:maction><m:mo>&#8805;</m:mo><m:maction actiontype="link" dsi:type="simple" dsi:href="#N"><m:mi mathcolor="#EE0000" mathvariant="bold">N</m:mi></m:maction><m:mo>+</m:mo><m:mn>1</m:mn></m:math>.
</div></dd><dt class="paramhead"><a name="N" id="N"/>3: &#160;&#160;&#8194; N &#8211; INTEGER<span class="pclass">Input</span></dt><dd><div class="paramtext"><i>On entry</i>: 

<m:math><m:mi>n</m:mi></m:math>, the order of the matrix <m:math><m:mi>A</m:mi></m:math>.</div><div class="paramtext"><i>Constraint</i>:
  <m:math><m:maction actiontype="link" dsi:type="simple" dsi:href="#N"><m:mi mathcolor="#EE0000" mathvariant="bold">N</m:mi></m:maction><m:mo>&#8805;</m:mo><m:mn>1</m:mn></m:math>.
</div></dd><dt class="paramhead"><a name="B" id="B"/>4: &#160;&#160;&#8194; B(<a class="arg" href="#LDB">LDB</a>,<a class="arg" href="#N">N</a>) &#8211; <span class="bitalic">double precision</span> array<span class="pclass">Output</span></dt><dd><div class="paramtext"><i>On exit</i>: the lower triangle of the inverse matrix <m:math><m:mi>X</m:mi></m:math>, with <m:math><m:msub><m:mi>x</m:mi><m:mrow><m:mi>i</m:mi><m:mi>j</m:mi></m:mrow></m:msub></m:math>&#160;stored in <m:math><m:mrow><m:maction actiontype="link" dsi:type="simple" dsi:href="#B"><m:mi mathcolor="#EE0000" mathvariant="bold">B</m:mi></m:maction><m:mfenced separators="," open="(" close=")"><m:mi>i</m:mi><m:mi>j</m:mi></m:mfenced></m:mrow></m:math>, for <m:math><m:mi>i</m:mi><m:mo>&#8805;</m:mo><m:mi>j</m:mi></m:math>.</div></dd><dt class="paramhead"><a name="LDB" id="LDB"/>5: &#160;&#160;&#8194; LDB &#8211; INTEGER<span class="pclass">Input</span></dt><dd><div class="paramtext"><i>On entry</i>: the first dimension of the array <a class="arg" href="#B">B</a> as declared in the (sub)program from which F01ABF is called.</div><div class="paramtext"><i>Constraint</i>:
  <m:math><m:maction actiontype="link" dsi:type="simple" dsi:href="#LDB"><m:mi mathcolor="#EE0000" mathvariant="bold">LDB</m:mi></m:maction><m:mo>&#8805;</m:mo><m:maction actiontype="link" dsi:type="simple" dsi:href="#N"><m:mi mathcolor="#EE0000" mathvariant="bold">N</m:mi></m:maction></m:math>.
</div></dd><dt class="paramhead"><a name="Z" id="Z"/>6: &#160;&#160;&#8194; Z(<a class="arg" href="#N">N</a>) &#8211; <span class="bitalic">double precision</span> array<span class="pclass">Workspace</span></dt><dt class="paramhead"><a name="IFAIL" id="IFAIL"/>7: &#160;&#160;&#8194; IFAIL &#8211; INTEGER<span class="pclass">Input/Output</span></dt><dd>
<div class="paramtext"><i>On entry</i>: <a class="arg" href="#IFAIL">IFAIL</a> must be set to <m:math><m:mn>0</m:mn></m:math>, <m:math><m:mrow><m:mo>-</m:mo><m:mn>1</m:mn></m:mrow><m:mtext>&#8203; or &#8203;</m:mtext><m:mn>1</m:mn></m:math>. If you are unfamiliar with this parameter you should refer to <a class="sec" href="../GENINT/essint.xml#library3">Section 3.3</a> in  the Essential Introduction for details.</div>
<div class="paramtext"><i>On exit</i>: <m:math><m:maction actiontype="link" dsi:type="simple" dsi:href="#IFAIL"><m:mi mathcolor="#EE0000" mathvariant="bold">IFAIL</m:mi></m:maction><m:mo>=</m:mo><m:maction actiontype="link" dsi:type="simple" dsi:href="#errors"><m:mn mathcolor="#003399" mathvariant="bold">0</m:mn></m:maction></m:math>&#160;unless the routine detects an error (see <a class="sec" href="#errors">Section 6</a>). <div class="paramtext">For environments where it might be inappropriate to halt program execution when an error is detected, the value <m:math><m:mrow><m:mo>-</m:mo><m:mn>1</m:mn></m:mrow><m:mtext>&#8203; or &#8203;</m:mtext><m:mn>1</m:mn></m:math>&#160;is recommended.  If the output of error messages is undesirable, then the value <m:math><m:mn>1</m:mn></m:math>&#160;is recommended.  Otherwise, if you are not familiar with this parameter, the recommended value is <m:math><m:mn>0</m:mn></m:math>.  <b>When the value <m:math><m:mrow><m:mo>-</m:mo><m:mn mathvariant="bold">1</m:mn></m:mrow><m:mtext>&#8203; or &#8203;</m:mtext><m:mn mathvariant="bold">1</m:mn></m:math>&#160;is used it is essential to test the value of <a class="arg" href="#IFAIL">IFAIL</a> on exit.</b></div></div></dd></dl><h2 class="standard"><a class="sec" name="errors" id="errors"/>6&#160;&#160;Error Indicators and Warnings</h2>
<div class="paramtext">If on entry <m:math><m:maction actiontype="link" dsi:type="simple" dsi:href="#IFAIL"><m:mi mathcolor="#EE0000" mathvariant="bold">IFAIL</m:mi></m:maction><m:mo>=</m:mo><m:maction actiontype="link" dsi:type="simple" dsi:href="#errors"><m:mn mathcolor="#003399" mathvariant="bold">0</m:mn></m:maction></m:math>&#160;or <m:math><m:maction actiontype="link" dsi:type="simple" dsi:href="#errors"><m:mn mathcolor="#003399" mathvariant="bold">-1</m:mn></m:maction></m:math>, explanatory error messages are output on the current error message unit (as defined by <a class="rout" href="../X04/x04aaf.xml">X04AAF</a>).</div><div class="paramtext">Errors or warnings detected by the routine:</div>
<dl class="ifail">
<dt class="errorhead"><a name="IFeq1" id="IFeq1"/><m:math><m:maction actiontype="link" dsi:type="simple" dsi:href="#IFAIL"><m:mi mathcolor="#EE0000" mathvariant="bold">IFAIL</m:mi></m:maction><m:mo>=</m:mo><m:mn>1</m:mn></m:math></dt>
<dd>
<div class="paramtext">The matrix <m:math><m:mi>A</m:mi></m:math>&#160;is not positive-definite, possibly due to rounding errors.</div>
</dd>
</dl><dl class="ifail">
<dt class="errorhead"><a name="IFeq2" id="IFeq2"/><m:math><m:maction actiontype="link" dsi:type="simple" dsi:href="#IFAIL"><m:mi mathcolor="#EE0000" mathvariant="bold">IFAIL</m:mi></m:maction><m:mo>=</m:mo><m:mn>2</m:mn></m:math></dt>
<dd>
<div class="paramtext">The refinement process fails to converge, i.e., the matrix <m:math><m:mi>A</m:mi></m:math>&#160;is ill-conditioned.</div>
</dd>
</dl><dl class="ifail">
<dt class="errorhead"><a name="IFeq3" id="IFeq3"/><m:math><m:maction actiontype="link" dsi:type="simple" dsi:href="#IFAIL"><m:mi mathcolor="#EE0000" mathvariant="bold">IFAIL</m:mi></m:maction><m:mo>=</m:mo><m:mn>3</m:mn></m:math></dt>
<dd>
<div class="paramtext"><m:math>
 <m:maction actiontype="link" dsi:type="simple" dsi:href="#N"><m:mi mathcolor="#EE0000" mathvariant="bold">N</m:mi></m:maction><m:mo>&lt;</m:mo><m:mn>1</m:mn>
</m:math>, or <m:math>
 <m:maction actiontype="link" dsi:type="simple" dsi:href="#LDA"><m:mi mathcolor="#EE0000" mathvariant="bold">LDA</m:mi></m:maction><m:mo>&lt;</m:mo><m:maction actiontype="link" dsi:type="simple" dsi:href="#N"><m:mi mathcolor="#EE0000" mathvariant="bold">N</m:mi></m:maction><m:mo>+</m:mo><m:mn>1</m:mn>
</m:math>, or <m:math>
 <m:maction actiontype="link" dsi:type="simple" dsi:href="#LDB"><m:mi mathcolor="#EE0000" mathvariant="bold">LDB</m:mi></m:maction><m:mo>&lt;</m:mo><m:maction actiontype="link" dsi:type="simple" dsi:href="#N"><m:mi mathcolor="#EE0000" mathvariant="bold">N</m:mi></m:maction>
</m:math>.</div>
</dd>
</dl><h2 class="standard"><a class="sec" name="accuracy" id="accuracy"/>7&#160;&#160;Accuracy</h2>
<div class="paramtext">The computed inverse should be correct to full machine accuracy.  For a detailed error analysis see page 40 of <a class="ref" href="#ref103">Wilkinson and Reinsch (1971)</a>.</div><h2 class="standard"><a class="sec" name="fcomments" id="fcomments"/>8&#160;&#160;Further Comments</h2>
<div class="paramtext">The time taken by F01ABF is approximately proportional to  <m:math><m:msup><m:mi>n</m:mi><m:mn>3</m:mn></m:msup></m:math>.</div><h2 class="standard"><a class="sec" name="example" id="example"/>9&#160;&#160;Example</h2>
<div class="paramtext">This example finds the inverse of the <m:math><m:mn>4</m:mn></m:math>&#160;by <m:math><m:mn>4</m:mn></m:math>&#160;matrix:

<div class="formula"><table class="formula"><tr><td class="formula"><m:math display="block">
 <m:mfenced><m:mtable columnalign="right">
  <m:mtr>
   <m:mtd><m:mn>5</m:mn></m:mtd>
   <m:mtd><m:mn>7</m:mn></m:mtd>
   <m:mtd><m:mn>6</m:mn></m:mtd>
   <m:mtd><m:mn>5</m:mn></m:mtd>
  </m:mtr><m:mtr>
   <m:mtd><m:mn>7</m:mn></m:mtd>
   <m:mtd><m:mn>10</m:mn></m:mtd>
   <m:mtd><m:mn>8</m:mn></m:mtd>
   <m:mtd><m:mn>7</m:mn></m:mtd>
  </m:mtr><m:mtr>
   <m:mtd><m:mn>6</m:mn></m:mtd>
   <m:mtd><m:mn>8</m:mn></m:mtd>
   <m:mtd><m:mn>10</m:mn></m:mtd>
   <m:mtd><m:mn>9</m:mn></m:mtd>
  </m:mtr><m:mtr>
   <m:mtd><m:mn>5</m:mn></m:mtd>
   <m:mtd><m:mn>7</m:mn></m:mtd>
   <m:mtd><m:mn>9</m:mn></m:mtd>
   <m:mtd><m:mn>10</m:mn></m:mtd>
  </m:mtr>
 </m:mtable></m:mfenced>
<m:mtext>.</m:mtext>
</m:math></td><td class="formula2"/></tr></table></div></div><h3 class="standard"><a class="sec" name="examtext" id="examtext"/>9.1&#160;&#160;Program Text</h3>
<p><a class="verbatimref" href="../../examples/source/f01abfe.f">Program Text (f01abfe.f)</a></p><h3 class="standard"><a class="sec" name="examdata" id="examdata"/>9.2&#160;&#160;Program Data</h3>
<p><a class="verbatimref" href="../../examples/data/f01abfe.d">Program&#160;Data (f01abfe.d)</a></p><h3 class="standard"><a class="sec" name="examresults" id="examresults"/>9.3&#160;&#160;Program Results</h3>
<p><a class="verbatimref" href="../../examples/baseresults/f01abfe.r">Program Results (f01abfe.r)</a></p>
<hr/><div><a class="rout" href="../../pdf/F01/f01abf.pdf">F01ABF (PDF version)</a></div><div><a class="chap" href="f01conts.xml">F01 Chapter Contents</a></div><div><a class="chapint" href="f01intro.xml">F01 Chapter Introduction</a></div>
<div><a class="htmltoc" href="../FRONTMATTER/manconts.xml">NAG Library Manual</a></div>
<div><hr/><a class="genint" href="../FRONTMATTER/copyright.xml">&#169; The Numerical Algorithms Group Ltd, Oxford, UK. 2009</a></div></body></html>
