rooted_trees
¶
- class nodepy.rooted_trees.RootedTree(strg)[source]¶
A rooted tree is a directed acyclic graph with one node, which has no incoming edges, designated as the root. Rooted trees are useful for analyzing the order conditions of multistage numerical ODE solvers, such as Runge-Kutta methods and other general linear methods.
The trees are represented as strings, using one of the notations introduced by Butcher (the third column of Table 300(I) of Butcher’s text). The character ‘T’ is used in place of \(\tau\) to represent a vertex, and braces ‘{ }’ are used instead of brackets ‘[ ]’ to indicate that everything inside the braces is joined to a single parent node. Thus the first four trees are:
‘T’, ‘{T}’, ‘{T^2}’, {{T}}’
These can be generated using the function
list_trees()
, which returns a list of all trees of a given order:>>> from nodepy import * >>> for p in range(4): print(rt.list_trees(p)) [''] ['T'] ['{T}'] ['{{T}}', '{T^2}']
Note that the tree of order 0 is indicated by an empty string.
If the tree contains an edge from vertex \(A\) to vertex \(B\), vertex \(B\) is said to be a child of vertex \(A\). A vertex with no children is referred to as a leaf.
Warning
One important convention is assumed in the code; namely, that at each level, leaves are listed first (before any other subtrees), and if there are \(n\) leaves, we write ‘T^n’.
Note
Currently, powers cannot be used for subtrees; thus
‘{{T}{T}}’
is valid, while
‘{{T}^2}’
is not. This restriction may be lifted in the future.
Examples:
>>> from nodepy import rooted_trees as rt >>> tree=rt.RootedTree('{T^2{T{T}}{T}}') >>> tree.order() 9 >>> tree.density() 144 >>> tree.symmetry() 2
Topologically equivalent trees are considered equal:
>>> tree2=RootedTree('{T^2{T}{T{T}}}') >>> tree2==tree True
We can generate Python code to evaluate the elementary weight corresponding to a given tree for a given class of methods:
>>> rk.elementary_weight_str(tree) 'dot(b,dot(A,c)*dot(A,c*dot(A,c))*c**2)'
- References:
- TODO: - Check validity of strg more extensively
Accept any leaf ordering, but convert it to our convention
convention for ordering of subtrees?
- order()[source]¶
The order of a rooted tree, denoted \(r(t)\), is the number of vertices in the tree.
Examples:
>>> from nodepy import rooted_trees as rt >>> tree=rt.RootedTree('{T^2{T{T}}}') >>> tree.order() 7
- density()[source]¶
The density of a rooted tree, denoted by \(\gamma(t)\), is the product of the orders of the subtrees.
Examples:
>>> from nodepy import rooted_trees as rt >>> tree=rt.RootedTree('{T^2{T{T}}}') >>> tree.density() 56
Reference: [But08] p. 127, eq. 301(c)
- symmetry()[source]¶
The symmetry \(\sigma(t)\) of a rooted tree is…
Examples:
>>> from nodepy import rooted_trees as rt >>> tree=rt.RootedTree('{T^2{T{T}}}') >>> tree.symmetry() 2
Reference: [But08] p. 127, eq. 301(b)
- Dmap()[source]¶
Butcher’s function \(D(t)\) which represents differentiation. Defined by \(D(t) = 0\) except for D(‘T’)=1.
Reference: [BT97]
- lamda(alpha, extraargs=[])[source]¶
Computes Butcher’s functional lambda on a given tree for the function alpha. This is used to compute the product of two functions on trees.
- Input:
alpha – a function on rooted trees
extraargs – a list containing any additional arguments that must be passed to alpha
- Output:
tprod – a list of trees [t1, t2, …]
fprod – a list of numbers [a1, a2, …]
The meaning of the output is that \(\lambda(\alpha, t)(\beta) = a_1 \beta(t_1) + a_2 \beta(t_2) + \cdots\).
Examples:
>>> from nodepy import rt >>> tree = rt.RootedTree('{T{T}}') >>> tree.lamda(rt.Emap) (['T', '{T}', '{{T}}', '{T}', '{T^2}', '{T{T}}'], [1/2, 1, 1, 1/2, 1, 1])
Reference: [But08] pp. 275-276
- lamda_str(alpha, extraargs=[])[source]¶
Alternate version of
lamda()
, but returns a string. Hopefully we can get rid of this (and the other string functions) when SAGE can handle noncommutative symbolic algebra.
- Gprod(alpha, beta, alphaargs=[], betaargs=[])[source]¶
Returns the product of two functions on a given tree.
- Input:
alpha, beta – two functions on rooted trees that return symbolic or numeric values
alphaargs – a string containing any additional arguments that must be passed to function alpha
betaargs – a string containing any additional arguments that must be passed to function beta
Output:
(alpha*beta)(self) – i.e., the function that is the product (in \(G\)) of the functions alpha and beta. Note that this product is not commutative.
The product is given by
$$ (\alpha*\beta)(‘’)=\beta(‘’) $$
$$ (\alpha*\beta)(t) = \lambda(\alpha,t)(\beta) + \alpha(t)\beta(‘’) $$
Note
Gprod()
can be used to compute products of more than two functions by passingGprod()
itself in as beta, and providing the remaining functions to be multiplied as betaargs.Examples:
>>> from nodepy import rt >>> tree = rt.RootedTree('{T{T}}') >>> tree.Gprod(rt.Emap,Dmap) 1/2
Reference: [But08] p. 276, Thm. 386A
- Gprod_str(alpha, beta, alphaargs=[], betaargs=[])[source]¶
Alternate version of
Gprod()
, but operates on strings. Hopefully can be eliminated later in favor of symbolic manipulation.
- plot(nrows=1, ncols=1, iplot=1, ttitle='')[source]¶
Plots the rooted tree.
- Input: (optional)
nrows, ncols – number of rows and columns of subplots in the figure
iplot – index of the subplot in which to plot this tree
These are only necessary if plotting more than one tree in a single figure using subplot.
Output: None.
The plot is created recursively by plotting the root, parsing the subtrees, plotting the subtrees’ roots, and calling
_plot_subtree
on each child
- list_equivalent_trees()[source]¶
Returns a list of all strings (subject to our assumptions) equivalent to a given tree
- Input:
self – any rooted tree
- Output:
treelist – a list of all the ‘legal’ tree strings that produce the same tree.
The list of equivalent trees is obtained by taking all permutations of the (non-leaf) subtrees. This routine is used to test equality of trees.
- nodepy.rooted_trees.plot_all_trees(p, title='str')[source]¶
Plots all rooted trees of order \(p\).
Example:
Plot all trees of order 4:
>>> from nodepy import rt >>> rt.plot_all_trees(4) <Figure...
- nodepy.rooted_trees.list_trees(p, ind='all')[source]¶
Returns rooted trees of order \(p\).
Input:
p – order of trees desired
ind – if given, returns a single tree corresponding to this index. Not very useful since the ordering isn’t obvious.
OUTPUT: list of all trees of order \(p\) (or just one, if ind is provided).
Generates the rooted trees using Albrecht’s ‘Recursion 3’.
Examples:
Produce column of Butcher’s Table 302(I):
>>> for i in range(1,11): ... forest=list_trees(i) ... print(len(forest)) 1 1 2 4 9 20 48 115 286 719
Warning
This code is complete only up to order 10. We need to extend it by adding more subloops for \(p>10\).
TODO: Implement Butcher’s formula (Theorem 302B) for the number of trees and determine to what order this is valid.
Reference: [Alb96]
- nodepy.rooted_trees.Dprod(tree, alpha)[source]¶
Evaluate (alpha*D)(t). Note that this is not equal to (D*alpha)(t). This function is necessary (rather than just using :meth:RootedTree.Gprod`) in order to avoid infinite recursions.
Examples:
>>> from nodepy import rt >>> tree = rt.RootedTree('{T{T}}') >>> Dprod(tree,Emap) 1/2
- nodepy.rooted_trees.Dmap(tree)[source]¶
Butcher’s function \(D(t)\). Represents differentiation. Defined by \(D(t)=0\) except for \(D('T')=1\).
- nodepy.rooted_trees.Gprod(tree, alpha, beta, alphaargs='', betaargs=[])[source]¶
Returns the product of two functions on a given tree.
See Butcher p. 276, Thm. 386A.
- nodepy.rooted_trees.Emap(tree, a=1)[source]¶
Butcher’s function \(E^a(t)\). Gives the \(B\)-series for the exact solution advanced ‘a’ steps in time.
Examples:
>>> from nodepy import rooted_trees as rt >>> tree=rt.RootedTree('{T^2{T{T}}}') >>> rt.Emap(tree) 1/56 >>> rt.Emap(tree,a=2) 16/7
Reference: [BT97]
- nodepy.rooted_trees.recursiveVectors(p, ind='all')[source]¶
Generate recursive vectors using Albrecht’s ‘recursion 1’. These are essentially the order conditions for Runge-Kutta methods, excluding those that correspond to bushy trees. More specifically, these are the vectors that must be orthogonal to the vector of weights b.
Note that the individual order conditions obtained from this algorithm are different from those obtained using Butcher’s approach. But as a set of conditions up to some order they are, of course, equivalent.
Follows [Alb96] p. 1718
Warning
This code is complete only up to order 14. We need to extend it by adding more subloops for \(p>14\).
Example
Count number of conditions for order 12:
>>> from nodepy import rt >>> v = rt.recursiveVectors(12) >>> print(len(v)) 4765