| Quellcodebibliothek Statistik Leitseite products/sources/formale Sprachen/GAP/doc/tut/ (Algebra von RWTH Aachen Version 4.15.1©) Datei vom 18.9.2025 mit Größe 18 kB |
|
Impressum opers.xml Sprache: XML |
|||||||||
|
<!-- %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% -->
<!-- %% --> <!-- %W opers.tex GAP documentation Heiko Theißen --> <!-- %% --> <!-- %% --> <!-- %Y Copyright 1997, Lehrstuhl D für Mathematik, RWTH Aachen, Germany --> <!-- %% --> <!-- %% This file contains a tutorial introduction to operations. --> <!-- %% --> <P/> <!-- %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% --> <Chapter Label="Operations and Methods"> <Heading>Operations and Methods</Heading> <!-- %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% --> <Section Label="Attributes"> <Heading>Attributes</Heading> In the preceding chapters, we have seen how to obtain information about mathematical objects in &GAP;: We have to pass the object as an argument to a function. For example, if <A>G</A> is a group one can call <C>Size( <A>G</A> )</C>, and the function will return a value, in our example an integer which is the size of <A>G</A>. Computing the size of a group generally requires a substantial amount of work, therefore it seems desirable to store the size somewhere once it has been calculated. You should imagine that &GAP; stores the size in some place associated with the object <A>G</A> when <C>Size( <A>G</A> )</C> is executed for the first time, and if this function call is executed again later, the size is simply looked up and returned, without further computation. <P/> <Index Subkey="of an attribute">getter</Index> <Index Subkey="of an attribute">setter</Index> <Index Subkey="of an attribute">tester</Index> <Index>methods</Index> This means that the behavior of the function <Ref Func="Size" BookName="ref"/> has to depend on whether the size for the argument <A>G</A> is already known, and if not, that the size must be stored after it has been calculated. These two extra tasks are done by two other functions that accompany <C>Size( <A>G</A> )</C>, namely the <E>tester</E> <C>HasSize( <A>G</A> )</C> and the <E>setter</E> <C>SetSize( <A>G</A>, <A>size</A> )</C>. The tester returns <K>true</K> or <K>false</K> according to whether <A>G</A> has already stored its size, and the setter puts <A>size</A> into a place from where <A>G</A> can directly look it up. The function <Ref Func="Size" BookName="ref"/> itself is called the <E>getter</E>, and from the preceding discussion we see that there must really be at least two <E>methods</E> for the getter: One method is used when the tester returns <K>false</K>; it is the method which first does the real computation and then executes the setter with the computed value. A second method is used when the tester returns <K>true</K>; it simply returns the stored value. This second method is also called the <E>system getter</E>. &GAP; functions for which several methods can be available are called <E>operations</E>, so <Ref Func="Size" BookName="ref"/> is an example of an operation. <P/> <Log><![CDATA[ gap> G := Group(List([1..3], i-> Random(SymmetricGroup(53))));; gap> Size( G ); time; # the time may of course vary on your machine 4274883284060025564298013753389399649690343788366813724672000000000000 196 gap> Size( G ); time; 4274883284060025564298013753389399649690343788366813724672000000000000 0 ]]></Log> <P/> The convenient thing for the user is that &GAP; automatically chooses the right method for the getter, i.e., it calls a real-work getter at most once and the system getter in all subsequent occurrences. <E>At most once</E> because the value of a function call like <C>Size( <A>G</A> )</C> can also be set for <A>G</A> before the getter is called at all; for example, one can call the setter directly if one knows the size. <P/> The size of a group is an example of a class of things which in &GAP; are called <E>attributes</E>. Every attribute in &GAP; is represented by a triple of a getter, a setter and a tester. When a new attribute is declared, all three functions are created together and the getter contains references to the other two. This is necessary because when the getter is called, it must first consult the tester, and perhaps execute the setter in the end. Therefore the getter could be implemented as follows: <Log><![CDATA[ getter := function( obj ) local value; if tester( obj ) then value := system_getter( obj ); else value := real_work_getter( obj ); setter( obj, value ); fi; return value; end; ]]></Log> The only function which depends on the mathematical nature of the attribute is the real-work getter, and this is of course what the programmer of an attribute has to install. In both cases, the getter returns the same value, which we also call the value of the attribute (properly: the value of the attribute for the object <C>obj</C>). By the way: The names for setter and tester of an attribute are always composed from the prefix <C>Set</C> resp. <C>Has</C> and the name of the getter. <P/> As a (not typical) example, note that the &GAP; function <Ref Func="Random" BookName="ref"/>, although it takes only one argument, is of course <E>not</E> an attribute, because otherwise the first random element of a group would be stored by the setter and returned over and over again by the system getter every time <Ref Func="Random" BookName="ref"/> is called in the sequel. <P/> There is a general important rule about attributes: <E>Once the value of an attribute for an object has been set, it cannot be reset, i.e., it cannot be changed any more.</E> This is achieved by having two methods not only for the getter but also for the setter: If an object already has an attribute value stored, i.e., if the tester returns <K>true</K>, the setter simply does nothing. <P/> <Example><![CDATA[ gap> G := SymmetricGroup(8);; Size(G); 40320 gap> SetSize( G, 0 ); Size( G ); 40320 ]]></Example> <P/> <E>Summary.</E> In this section we have introduced attributes as triples of getter, setter and tester and we have explained how these three functions work together behind the scene to provide automatic storage and look-up of values that have once been calculated. We have seen that there can be several methods for the same function among which &GAP; automatically selects an appropriate one. </Section> <!-- %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% --> <Section Label="Properties and Filters"> <Heading>Properties and Filters</Heading> <Index>filters</Index> <Index Subkey="selection">methods</Index> Certain attributes, like <Ref Func="IsAbelian" BookName="ref"/>, are boolean-valued. Such attributes are known to &GAP; as <E>properties</E>, because their values are stored in a slightly different way. A property also has a getter, a setter and a tester, but in this case, the getter as well as the tester returns a boolean value. Therefore &GAP; stores both values in the same way, namely as bits in a boolean list, thereby treating property getters and all testers (of attributes or properties) uniformly. These boolean-valued functions are called <E>filters</E>. You can imagine a filter as a switch which is set either to <K>true</K> or to <K>false</K>. For every &GAP; object there is a boolean list which has reserved a bit for every filter &GAP; knows about. Strictly speaking, there is one bit for every <E>simple filter</E>, and these simple filters can be combined with <C>and</C> to form other filters (which are then <K>true</K> if and only if all the corresponding bits are set to <K>true</K>). For example, the filter <C>IsPermGroup and IsSolvableGroup</C> is made up from several simple filters. <P/> Since they allow only two values, the bits which represent filters can be compared very quickly, and the scheme by which &GAP; chooses the method, e.g., for a getter or a setter (as we have seen in the previous section), is mostly based on the examination of filters, not on the examination of other attribute values. Details of this <E>method selection</E> are described in chapter <Ref Chap="Method Selection" BookName="ref"/>. <P/> We only present the following rule of thumb here: Each installed method for an attribute, for example <Ref Func="Size" BookName="ref"/>, has a <Q>required filter</Q>, which is made up from certain simple filters which must yield <K>true</K> for the argument <A>obj</A> for this method to be applicable. To execute a call of <C>Size( <A>obj</A> )</C>, &GAP; selects among all applicable methods the one whose required filter combines the most simple filters; the idea behind is that the more an algorithm requires of <A>obj</A>, the more efficient it is expected to be. For example, if <A>obj</A> is a permutation group that is not (known to be) solvable, a method with required filter <C>IsPermGroup and IsSolvableGroup</C> is not applicable, whereas a method with required filter <Ref Func="IsPermGroup" BookName="ref"/> can be chosen. On the other hand, if <A>obj</A> was known to be solvable, the method with required filter <C>IsPermGroup and IsSolvableGroup</C> would be preferred to the one with required filter <Ref Func="IsPermGroup" BookName="ref"/>. <P/> It may happen that a method is applicable for a given argument but cannot compute the desired value. In such cases, the method will execute the statement <C>TryNextMethod();</C>, <Index Key="TryNextMethod"><C>TryNextMethod</C></Index> and &GAP; calls the next applicable method. For example, <Cite Key="Sims90b"/> describes an algorithm to compute the size of a solvable permutation group, which can be used also to decide whether or not a permutation group is solvable. Suppose that the function <C>size_solvable</C> implements this algorithm, and that is returns the order of the group if it is solvable and <K>fail</K> otherwise. Then we can install the following method for <Ref Func="Size" BookName="ref"/> with required filter <Ref Func="IsPermGroup" BookName="ref"/>. <Log><![CDATA[ function( G ) local value; value := size_solvable( G ); if value <> fail then return value; else TryNextMethod(); fi; end; ]]></Log> This method can then be tried on every permutation group (whether known to be solvable or not), and it would include a mandatory solvability test. <P/> If no applicable method (or no next applicable method) is found, &GAP; stops with an error message of the form <Log><![CDATA[ Error, no method found! For debugging hints type ?Recovery from NoMethodFound Error, no 1st choice method found for `Size' on 1 arguments called from ... lines deleted here ... ]]></Log> <P/> You would get an error message as above if you asked for <C>Size( 1 )</C>. The message simply says that there is no method installed for calculating the size of <C>1</C>. Section <Ref Sect="Recovery from NoMethodFound-Errors" BookName="ref"/> contains more information on how to deal with these messages. <P/> <E>Summary.</E> In this section we have introduced properties as special attributes, and filters as the general concept behind property getters and attribute testers. The values of the filters of an object govern how the object is treated in the selection of methods for operations. </Section> <!-- %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% --> <Section Label="Immediate and True Methods"> <Heading>Immediate and True Methods</Heading> <Index Subkey="immediate">methods</Index> <Index Subkey="true">methods</Index> In the example in Section <Ref Sect="Properties and Filters"/>, we have mentioned that the operation <Ref Func="Size" BookName="ref"/> has a method for solvable permutation groups that is so far superior to the method for general permutation groups that it seems worthwhile to try it even if nothing is known about solvability of the group of which the <Ref Func="Size" BookName="ref"/> is to be calculated. There are other examples where certain methods are even <Q>cheaper</Q> to execute. For example, if the size of a group is known it is easy to check whether it is odd, and if so, the Feit-Thompson theorem allows us to set <Ref Func="IsSolvableGroup" BookName="ref"/> to <K>true</K> for this group. &GAP; utilizes this celebrated theorem by having an <E>immediate method</E> for <Ref Func="IsSolvableGroup" BookName="ref"/> with required filter <C>HasSize</C> which checks parity of the size and either sets <Ref Func="IsSolvableGroup" BookName="ref"/> or does nothing, i.e., calls <C>TryNextMethod()</C>. These immediate methods are executed automatically for an object whenever the value of a filter changes, so solvability of a group will automatically be detected when an odd size has been calculated for it (and therefore the value of <C>HasSize</C> for that group has changed to <K>true</K>). <P/> Some methods are even more immediate, because they do not require any calculation at all: They allow a filter to be set if another filter is also set. In other words, they model a mathematical implication like <C>IsGroup and IsCyclic</C> implies <C>IsSolvableGroup</C> and such implications can be installed in &GAP; as <E>true methods</E>. To execute true methods, &GAP; only needs to do some bookkeeping with its filters, therefore true methods are much faster than immediate methods. <P/> How immediate and true methods are installed is described in <Ref Sect="Immediate Methods" BookName="ref"/> and <Ref Sect="Logical Implications" BookName="ref"/>. </Section> <!-- %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% --> <Section Label="Operations and Method Selection"> <Heading>Operations and Method Selection</Heading> <Index>operations</Index> The method selection is not only used to select methods for attribute getters but also for arbitrary <E>operations</E>, which can have more than one argument. In this case, there is a required filter for each argument (which must yield <K>true</K> for the corresponding arguments). <P/> Additionally, a method with at least two arguments may require a certain relation between the arguments, which is expressed in terms of the <E>families</E> of the arguments. For example, the methods for <C>ConjugateGroup( <A>grp</A>, <A>elm</A> )</C> require that <A>elm</A> lies in the family of elements from which <A>grp</A> is made, i.e., that the family of <A>elm</A> equals the <Q>elements family</Q> of <A>grp</A>. <P/> For permutation groups, the situation is quite easy: all permutations form one family, <Ref Var="PermutationsFamily" BookName="ref"/>, and each collection of permutations, for example each permutation group, each coset of a permutation group, or each dense list of permutations, lies in <C>CollectionsFamily( PermutationsFamily )</C>. <P/> For other kinds of group elements, the situation can be different. Every call of <Ref Func="FreeGroup" BookName="ref"/> constructs a new family of free group elements. &GAP; refuses to compute <C>One( FreeGroup( 1 ) ) * One( FreeGroup( 1 ) )</C> because the two operands of the multiplication lie in different families and no method is installed for this case. <P/> For further information on family relations, see <Ref Sect="Families" BookName="ref"/>. <P/> <Index Key="KnownPropertiesOfObject"><C>KnownPropertiesOfObject</C></Index> <Index Key="KnownTruePropertiesOfObject"><C>KnownTruePropertiesOfObject</C></Index> <Index Key="KnownAttributesOfObject"><C>KnownAttributesOfObject</C></Index> If you want to know which properties are already known for an object <A>obj</A>, or which properties are known to be true, you can use the functions <C>KnownPropertiesOfObject( <A>obj</A> )</C> resp. <C>KnownTruePropertiesOfObject( <A>obj</A> )</C>. This will print a list of names of properties. These names are also the identifiers of the property getters, by which you can retrieve the value of the properties (and confirm that they are really <K>true</K>). Analogously, there is the function <Ref Func="KnownAttributesOfObject" BookName="ref"/> which lists the names of the known attributes, leaving out the properties. <P/> <Index Key="ApplicableMethod"><C>ApplicableMethod</C></Index> Since &GAP; lets you know what it already knows about an object, it is only natural that it also lets you know what methods it considers applicable for a certain method, and in what order it will try them (in case <C>TryNextMethod()</C> occurs). <C>ApplicableMethod( <A>opr</A>, [ arg_1, arg_2, ... ] )</C> returns the first applicable method for the call <C><A>opr</A>( arg_1, arg_2, ... )</C>. More generally, <C>ApplicableMethod( <A>opr</A>, [ ... ], 0, <A>nr</A> )</C> returns the <A>nr</A>th applicable method (i.e., the one that would be chosen after <M><A>nr</A>-1</M> calls of <C>TryNextMethod</C>) and if <A>nr</A><C> = "all"</C>, the sorted list of all applicable methods is returned. For details, see <Ref Sect="Applicable Methods and Method Selection" BookName="ref"/>. <P/> <Index Key="TraceMethods"><C>TraceMethods</C></Index> If you want to see which methods are chosen for certain operations while &GAP; code is being executed, you can call the function <Ref Func="TraceMethods" BookName="ref" Label="for a list of operations"/> with a list of these operations as arguments. <P/> <Log><![CDATA[ gap> TraceMethods( [ Size ] ); gap> g:= Group( (1,2,3), (1,2) );; Size( g ); #I Size: for a permutation group #I Setter(Size): system setter #I Size: system getter #I Size: system getter 6 ]]></Log> <P/> The system getter is called once to fetch the freshly computed value for returning to the user. The second call is triggered by an immediate method. To find out by which, we can trace the immediate methods by saying <C>TraceImmediateMethods( true )</C>. <P/> <Log><![CDATA[ gap> TraceImmediateMethods( true ); gap> g:= Group( (1,2,3), (1,2) );; #I immediate: Size #I immediate: IsCyclic #I immediate: IsCommutative #I immediate: IsTrivial gap> Size( g ); #I Size: for a permutation group #I immediate: IsNonTrivial #I immediate: Size #I immediate: IsFreeAbelian #I immediate: IsTorsionFree #I immediate: IsNonTrivial #I immediate: GeneralizedPcgs #I Setter(Size): system setter #I Size: system getter #I immediate: IsPerfectGroup #I Size: system getter #I immediate: IsEmpty 6 gap> TraceImmediateMethods( false ); gap> UntraceMethods( [ Size ] ); ]]></Log> <P/> The last two lines switch off tracing again. We now see that the system getter was called by the immediate method for <Ref Func="IsPerfectGroup" BookName="ref"/>. Also the above-mentioned immediate method for <Ref Func="IsSolvableGroup" BookName="ref"/> was not used because the solvability of <C>g</C> was already found out during the size calculation (cf. the example in Section <Ref Sect="Properties and Filters"/>). <P/> <E>Summary.</E> In this section and the last we have looked some more behind the scenes and seen that &GAP; automatically executes immediate and true methods to deduce information about objects that is cheaply available. We have seen how this can be supervised by tracing the methods. </Section> </Chapter> <!-- %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% --> <!-- %% --> <!-- %E --> ¤ Dauer der Verarbeitung: 0.30 Sekunden (vorverarbeitet) ¤*© Formatika GbR, Deutschland |
|
