<?xml version="1.0" encoding="utf-8" standalone="no"?>
<!DOCTYPE document PUBLIC "-//CNX//DTD CNXML 0.5 plus MathML//EN" "http://cnx.rice.edu/cnxml/0.5/DTD/cnxml_mathml.dtd">
<document xmlns="http://cnx.rice.edu/cnxml" xmlns:md="http://cnx.rice.edu/mdml/0.4" xmlns:m="http://www.w3.org/1998/Math/MathML" xmlns:bib="http://bibtexml.sf.net/" id="new0">
  <name>A Hierarchy of Detail in the Haar System</name>
  <metadata>
  <md:version>2.3</md:version>
  <md:created>2003/01/17</md:created>
  <md:revised>2005/11/17 16:07:33.158 US/Central</md:revised>
  <md:authorlist>
      <md:author id="schniter">
      <md:firstname>Phil</md:firstname>
      
      <md:surname>Schniter</md:surname>
      <md:email>schniter@ee.eng.ohio-state.edu</md:email>
    </md:author>
  </md:authorlist>

  <md:maintainerlist>
    <md:maintainer id="charlet">
      <md:firstname>Charlet</md:firstname>
      
      <md:surname>Reedstrom</md:surname>
      <md:email>charlet@rice.edu</md:email>
    </md:maintainer>
    <md:maintainer id="schniter">
      <md:firstname>Phil</md:firstname>
      
      <md:surname>Schniter</md:surname>
      <md:email>schniter@ee.eng.ohio-state.edu</md:email>
    </md:maintainer>
  </md:maintainerlist>
  
  <md:keywordlist>
    <md:keyword>Haar</md:keyword>
    <md:keyword>scaling function</md:keyword>
  </md:keywordlist>

  <md:abstract/>
</metadata>


  <content>
    <para id="hp1">
      Given a mother scaling function
      <m:math>
	<m:apply>
	  <m:in/>
	  <m:apply>
	    <m:ci type="fn">φ</m:ci>
	    <m:ci>t</m:ci>
	  </m:apply>
	  <m:ci type="set"><m:msub> 
	      <m:mi>ℒ</m:mi> 
	      <m:mn>2</m:mn>
	    </m:msub></m:ci>
	</m:apply>
      </m:math>
      — the choice of which will be discussed later <!-- good
      place for cnxn, but to where?--> — let us construct
      scaling functions at
      "coarseness-level-<m:math><m:ci>k"</m:ci></m:math> and
      "shift-<m:math><m:ci>n</m:ci></m:math>" as follows:
      <m:math display="block">
	<m:apply>
	  <m:eq/>
	  <m:apply>
	    <m:ci type="fn"><m:msub>
		<m:mi>φ</m:mi>
		<m:mrow>
		  <m:mi>k</m:mi>
		  <m:mo>,</m:mo>
		  <m:mi>n</m:mi>
		</m:mrow>
	      </m:msub></m:ci>
	    <m:ci>t</m:ci>
	  </m:apply>
	  <m:apply>
	    <m:times/>
	    <m:apply>
	      <m:power/>
	      <m:cn>2</m:cn>
	      <m:apply>
		<m:minus/>
		<m:apply>
		  <m:divide/>
		  <m:ci>k</m:ci>
		  <m:cn>2</m:cn>
		</m:apply>
	      </m:apply>
	    </m:apply>
	    <m:apply>
	      <m:ci type="fn">φ</m:ci>
	      <m:apply>
		<m:minus/>
		<m:apply>
		  <m:times/>
		  <m:apply>
		    <m:power/>
		    <m:cn>2</m:cn>
		    <m:apply>
		      <m:minus/>
		      <m:ci>k</m:ci>
		    </m:apply>
		  </m:apply>
		  <m:ci>t</m:ci>
		</m:apply>
		<m:ci>n</m:ci>
	      </m:apply>
	    </m:apply>
	  </m:apply>
	</m:apply>
	<m:mtext>.</m:mtext>
      </m:math>
      Let us then use
      <m:math>
	<m:ci><m:msub>
	    <m:mi>V</m:mi>
	    <m:mi>k</m:mi>
	  </m:msub></m:ci>
      </m:math>
      to denote the subspace defined by linear combinations of
      scaling functions at the 
      <m:math>
	<m:ci><m:msup>
	    <m:mi>k</m:mi>
	    <m:mi>th</m:mi>
	  </m:msup></m:ci>
      </m:math>
      level:  

      <m:math display="block">
	<m:apply>
	  <m:eq/>
	  <m:ci><m:msub>
	      <m:mi>V</m:mi>
	      <m:mi>k</m:mi>
	    </m:msub></m:ci>
	  <m:apply>
	    <m:ci type="fn">span</m:ci>
	    <m:set>
	      <m:bvar>
		<m:apply>
		  <m:ci type="fn"><m:msub>
		      <m:mi>φ</m:mi>
		      <m:mrow>
			<m:mi>k</m:mi>
			<m:mo>,</m:mo>
			<m:mi>n</m:mi>
		      </m:mrow>
		    </m:msub></m:ci>
		  <m:ci>t</m:ci>
		</m:apply>
	      </m:bvar>
	      <m:condition>
		<m:apply>
		  <m:in/>
		  <m:ci>n</m:ci>
		  <m:integers/>
		</m:apply>
	      </m:condition>
	    </m:set>
	  </m:apply>
	</m:apply>
	<m:mtext>.</m:mtext>
      </m:math>
      In the Haar system, for example,    
      <m:math>
	<m:ci><m:msub>
	    <m:mi>V</m:mi>
	    <m:mn>0</m:mn>
	  </m:msub></m:ci>
      </m:math>
      and   
      <m:math>
	<m:ci><m:msub>
	    <m:mi>V</m:mi>
	    <m:mn>1</m:mn>
	  </m:msub></m:ci>
      </m:math>
      consist of signals with the characteristics of
      <m:math>
	<m:apply>
	  <m:ci type="fn"><m:msub>
	      <m:mi>x</m:mi>
	      <m:mn>0</m:mn>
	    </m:msub></m:ci>
	  <m:ci>t</m:ci>
	</m:apply>
      </m:math>
      and   
      <m:math>
	<m:apply>
	  <m:ci type="fn"><m:msub>
	      <m:mi>x</m:mi>
	      <m:mn>1</m:mn>
	    </m:msub></m:ci>
	  <m:ci>t</m:ci>
	</m:apply>
      </m:math> illustrated in <cnxn target="figb" strength="9"/>.
    </para>

    <figure id="figb">
      <media type="image/png" src="sq_scalingfn.png"/>
    </figure>

    <para id="hp2">
      We will be careful to choose a scaling function
      <m:math>
	<m:apply>
	  <m:ci type="fn">φ</m:ci>
	  <m:ci>t</m:ci>
	</m:apply>
      </m:math> which ensures that the following nesting property
      is satisfied:
      <m:math display="block">
	<m:apply>
	  <m:prsubset/>
	  <m:ci>…</m:ci>
	  <m:ci><m:msub>
	      <m:mi>V</m:mi>
	      <m:mn>2</m:mn>
	    </m:msub></m:ci>
	  <m:ci><m:msub>
	      <m:mi>V</m:mi>
	      <m:mn>1</m:mn>
	    </m:msub></m:ci>
	  <m:ci><m:msub>
	      <m:mi>V</m:mi>
	      <m:mn>0</m:mn>
	    </m:msub></m:ci>
	  <m:ci><m:msub>
	      <m:mi>V</m:mi>
	      <m:mn>-1</m:mn>
	    </m:msub></m:ci>
	  <m:ci><m:msub>
	      <m:mi>V</m:mi>
	      <m:mn>-2</m:mn>
	    </m:msub></m:ci>
	  <m:ci>…</m:ci>
	</m:apply>
      </m:math> 
      <m:math display="block">
	<m:mtext>coarse</m:mtext>
	<m:mo></m:mo>
	<m:mtext>              </m:mtext>
	<m:mo></m:mo>
	<m:mtext>detailed</m:mtext>
      </m:math>
      In other words, any signal in
      <m:math>
	<m:ci><m:msub>
	    <m:mi>V</m:mi>
	    <m:mi>k</m:mi>
	  </m:msub></m:ci>
      </m:math>
      can be constructed as a linear combination of <emphasis>more
	detailed</emphasis> signals in
      <m:math>
	<m:ci><m:msub>
	    <m:mi>V</m:mi>
	    <m:mrow>
	      <m:mi>k</m:mi>
	      <m:mo>−</m:mo>
	      <m:mn>1</m:mn>
	    </m:mrow>
	  </m:msub></m:ci>
      </m:math>
      .  (The Haar system gives proof that at least one such
      <m:math>
	<m:apply>
	  <m:ci type="fn">φ</m:ci>
	  <m:ci>t</m:ci>
	</m:apply>
      </m:math> exists.)
    </para>

    <para id="para3a">
      The nesting property can be depicted using the set-theoretic
      diagram, <cnxn target="vspacesfig" strength="9"/>, where
      <m:math>
	<m:ci><m:msub>
	    <m:mi>V</m:mi>
	    <m:mrow>
	      <m:mo>−</m:mo>
	      <m:mn>1</m:mn>
	    </m:mrow>
	  </m:msub></m:ci>
      </m:math>
      is represented by the contents of the largest egg (which
      includes the smaller two eggs),
      <m:math>
	<m:ci><m:msub>
	    <m:mi>V</m:mi>
	    <m:mn>0</m:mn>
	  </m:msub></m:ci>
      </m:math>
      is represented by the contents of the medium-sized egg (which
      includes the smallest egg), and 
      <m:math>
	<m:ci><m:msub>
	    <m:mi>V</m:mi>
	    <m:mn>1</m:mn>
	  </m:msub></m:ci>
      </m:math>
      is represented by the contents of the smallest egg.
    </para>

    <figure id="vspacesfig">
      <media type="image/png" src="vspaces.png"/>
    </figure>

    <para id="hp3">
      Going further, we will assume that
      <m:math>
	<m:apply>
	  <m:ci type="fn">φ</m:ci>
	  <m:ci>t</m:ci>
	</m:apply>
      </m:math> is designed to yield the following three important
      properties:  

      <list id="spaceex" type="enumerated">
	<item>
	  <m:math>
	    <m:set>
	      <m:bvar>
		<m:apply>
		  <m:ci type="fn"><m:msub>
		      <m:mi>φ</m:mi>
		      <m:mrow>
			<m:mi>k</m:mi>
			<m:mo>,</m:mo>
			<m:mi>n</m:mi>
		      </m:mrow>
		    </m:msub></m:ci>
		  <m:ci>t</m:ci>
		</m:apply>
	      </m:bvar>
	      <m:condition>
		<m:apply>
		  <m:in/>
		  <m:ci>n</m:ci>
		  <m:integers/>
		</m:apply>
	      </m:condition>
	    </m:set>
	  </m:math> constitutes an orthonormal basis for 
	  <m:math>
	    <m:ci><m:msub>
		<m:mi>V</m:mi>
		<m:mi>k</m:mi>
	      </m:msub></m:ci>
	  </m:math>,
	</item>
	<item><m:math>
	    <m:apply>
	      <m:eq/>
	      <m:ci><m:msub>
		  <m:mi>V</m:mi>
		  <m:mi>∞</m:mi>
		</m:msub></m:ci>
	      <m:set>
		<m:cn type="vector">0</m:cn>
	      </m:set>
	    </m:apply>
	  </m:math> (contains no signals).
	  <note type="footnote">
	    While at first glance it might seem that
	    <m:math>
	      <m:ci><m:msub>
		  <m:mi>V</m:mi>
		  <m:mi>∞</m:mi>
		</m:msub></m:ci>
	    </m:math>
	    should contain non-zero constant signals (<foreign>e.g.</foreign>, 
	    <m:math>
	      <m:apply>
		<m:eq/>
		<m:apply>
		  <m:ci type="fn">x</m:ci>
		  <m:ci>t</m:ci>
		</m:apply>
		<m:ci>a</m:ci>
	      </m:apply>
	    </m:math> for 
	    <m:math>
	      <m:apply>
		<m:in/>
		<m:ci>a</m:ci>
		<m:reals/>
	      </m:apply>
	    </m:math>), the only constant signal in
	    <m:math>
	      <m:ci type="set"><m:msub> 
		  <m:mi>ℒ</m:mi> 
		  <m:mn>2</m:mn>
		</m:msub></m:ci>
	    </m:math>
	    , the space of square-integrable signals, is the zero
	    signal.</note>
	</item>
	<item><m:math>
	    <m:apply>
	      <m:eq/>
	      <m:ci><m:msub>
		  <m:mi>V</m:mi>
		  <m:mrow>
		    <m:mo>−</m:mo>
		    <m:mi>∞</m:mi>
		  </m:mrow>
		</m:msub></m:ci>
	      <m:ci type="set"><m:msub> 
		  <m:mi>ℒ</m:mi> 
		  <m:mn>2</m:mn>
		</m:msub></m:ci>
	    </m:apply>
	  </m:math> (contains all signals).
	</item>
      </list>

      Because 
      <m:math>
	<m:set>
	  <m:bvar>
	    <m:apply>
	      <m:ci type="fn"><m:msub>
		  <m:mi>φ</m:mi>
		  <m:mrow>
		    <m:mi>k</m:mi>
		    <m:mo>,</m:mo>
		    <m:mi>n</m:mi>
		  </m:mrow>
		</m:msub></m:ci>
	      <m:ci>t</m:ci>
	    </m:apply>
	  </m:bvar>
	  <m:condition>
	    <m:apply>
	      <m:in/>
	      <m:ci>n</m:ci>
	      <m:integers/>
	    </m:apply>
	  </m:condition>
	</m:set>
      </m:math> is an orthonormal basis, the best (in
      <m:math>
	<m:ci type="set"><m:msub> 
	    <m:mi>ℒ</m:mi> 
	    <m:mn>2</m:mn>
	  </m:msub></m:ci>
      </m:math>
      norm) approximation of 
      <m:math>
	<m:apply>
	  <m:in/>
	  <m:apply>
	    <m:ci type="fn">x</m:ci>
	    <m:ci>t</m:ci>
	  </m:apply>
	  <m:ci type="set"><m:msub> 
	      <m:mi>ℒ</m:mi> 
	      <m:mn>2</m:mn>
	    </m:msub></m:ci>
	</m:apply>
      </m:math> at coarseness-level-<m:math><m:ci>k</m:ci></m:math> is
      given by the orthogonal projection, <cnxn target="projection" strength="9"/>
      <equation id="heq1">
	<m:math display="block">
	  <m:apply>
	    <m:eq/>
	    <m:apply>
	      <m:ci type="fn"><m:msub>
		  <m:mi>x</m:mi>
		  <m:mi>k</m:mi>
		</m:msub></m:ci>
	      <m:ci>t</m:ci>
	    </m:apply>
	    <m:apply>
	      <m:sum/>
	      <m:bvar><m:ci>n</m:ci></m:bvar>
	      <m:lowlimit>
		<m:apply>
		  <m:minus/>
		  <m:infinity/>
		</m:apply>
	      </m:lowlimit>
	      <m:uplimit>
		<m:infinity/>
	      </m:uplimit>
	      <m:apply>
		<m:times/>
		<m:ci><m:msub>
		    <m:mi>c</m:mi>
		    <m:mrow>
		      <m:mi>k</m:mi>
		      <m:mo>,</m:mo>
		      <m:mi>n</m:mi>
		    </m:mrow>
		  </m:msub></m:ci>
		<m:apply>
		  <m:ci type="fn"><m:msub>
		      <m:mi>φ</m:mi>
		      <m:mrow>
			<m:mi>k</m:mi>
			<m:mo>,</m:mo>
			<m:mi>n</m:mi>
		      </m:mrow>
		    </m:msub></m:ci>
		  <m:ci>t</m:ci>
		</m:apply>
	      </m:apply>
	    </m:apply>
	  </m:apply>
	</m:math>
      </equation>

      <equation id="heq2">
	<m:math display="block">
	  <m:apply>
	    <m:eq/>
	    <m:ci><m:msub>
		<m:mi>c</m:mi>
		<m:mrow>
		  <m:mi>k</m:mi>
		  <m:mo>,</m:mo>
		  <m:mi>n</m:mi>
		</m:mrow>
	      </m:msub></m:ci>
	    <m:apply>
	      <m:scalarproduct/>
	      <m:apply>
		<m:ci type="fn"><m:msub>
		    <m:mi>φ</m:mi>
		    <m:mrow>
		      <m:mi>k</m:mi>
		      <m:mo>,</m:mo>
		      <m:mi>n</m:mi>
		    </m:mrow>
		  </m:msub></m:ci>
		<m:ci>t</m:ci>
	      </m:apply>
	      <m:apply>
		<m:ci type="fn">x</m:ci>
		<m:ci>t</m:ci>
	      </m:apply>
	    </m:apply>
	  </m:apply>
	</m:math>
      </equation>
    </para>

    <figure id="projection">
      <media type="image/png" src="projection.png"/>
    </figure>

    <para id="hp4">
      We will soon derive conditions on the scaling function
      <m:math>
	<m:apply>
	  <m:ci type="fn">φ</m:ci>
	  <m:ci>t</m:ci>
	</m:apply>
      </m:math> which ensure that the properties above are
      satisfied.
    </para>
  </content>
  
</document>
