NAME
Math::NumSeq::BalancedBinary  balanced 1,0 bits
SYNOPSIS
use Math::NumSeq::BalancedBinary;
my $seq = Math::NumSeq::BalancedBinary>new;
my ($i, $value) = $seq>next;
DESCRIPTION
This sequence is integers with 1bits and 0bits balanced like opening and closing parentheses.
2, 10, 12, 42, 44, 50, 52, 56, 170, 172, 178, ...
starting i=1
Written in binary a 1bit is an opening "(" and a 0bit is a closing ")".
i value in parens
binary
  
1 10 ()
2 1010 () ()
3 1100 (())
4 101010 () () ()
5 101100 () (())
6 110010 (()) ()
7 110100 (() ())
8 111000 ((()))
9 10101010 () () () ()
10 10101100 () () (())
Balanced means the total number of 1s and 0s are the same and reading from high to low has count(1s) >= count(0s) at all times, ie. any closing ")" must have a preceding matching open "(".
Because the number of 1s and 0s are equal the width is always an even 2*w. The number of values with a given width is the Catalan number C(w) = (2w)!/(w!*(w+1)!). For example 6bit values w=6/2=3 gives C(3) = (2*3)!/(3!*4!) = 5 many such values, being i=4 through i=8 inclusive as shown above.
Binary Trees
The sequence values correspond to binary trees where each node can have a left and/or right child. Such a tree can be encoded by writing
0 if no node (empty tree)
1,lefttree,righttree at a node
The "lefttree" and "righttree" parts are the left and right legs from the node written out recursively. If those legs are both empty (ie. the node is a leaf) then they're empty trees and are "0" giving node "100". Otherwise the node is 1 followed by various more 1s and 0s. For example,
d
/
b c => 11001010 [0]
\ / ab c d ^final zero of encoding omitted
a
At "a" write 1 and recurse to write its left then right legs. The left leg is "b" so write 1 and the two legs of "b" are empty so write 0,0. That completes the left side of "a" so resume at the right side of "a" which is 1 for "c" and descend to the left and right of "c". The left of "c" is empty so write 0. The right of "c" is "d" so write 1 and the two empty legs of "d" are 0,0. The very final 0 from that rightmost leaf "d" is dropped (shown "[0]" above).
This encoding can be applied breadthfirst too by pushing the left and right descents onto a queue of pending work instead of onto a stack by recursing. In both cases there's an extra final 0 which is dropped. This 0 arises because in any binary tree with K nodes there are K+1 empty legs. That would give K many 1bits and K+1 many 0bits.
In this encoding the balanced binary condition "count 1s >= count 0s" corresponds to there being at least one unfinished node at any time in the traversal (by whichever node order).
The NumSeq
code here acts on values as numbers. Tree encodings like this are probably better handled as a string or list of bits.
Mountain Ranges
A cute interpretation of the opens and closes is as up and down slopes of a mountain range. 1bit for up, 0bit for down. For example,
/\
/ \ /\
/ \/ \
/ \/\

11110001100010
The mountain range must end at its starting level and must remain at or above its starting level at all times.
Numerical order of the values means narrower mountain ranges are before wider ones, and two ranges with equal width are ordered by downslope preceding upslope at the first bit difference.
FUNCTIONS
See "FUNCTIONS" in Math::NumSeq for behaviour common to all sequence classes.
Random Access
$value = $seq>ith($i)

Return the
$i
'th balanced binary number. $bool = $seq>pred($value)

Return true if
$value
is balanced binary. $i = $seq>value_to_i($value)

If
$value
is balanced binary then return its index i. If$value
is not balanced binary then returnundef
. $i = $seq>value_to_i_ceil($value)
$i = $seq>value_to_i_floor($value)

Return the index i of
$value
or if$value
is not a balanced binary integer then return the i of the next higher or lower value, respectively.
FORMULAS
Next
When stepping to the next value the number of 1s and 0s does not change. The 1s move to make a numerically higher value. The simplest is an isolated low 1bit. It must move up one place. For example,
11100100 isolated low 1bit
> shifts up
11101000
If the low 1 has a 1 above it then that bit must move up and the lower one goes to the low end of the value. For example
1110011000 pair of bits
> one shifts up, other drops to low end
1110100010
In general the lowest run of 1bits is changed to have the highest of them move up one place and the rest move down to be a ...101010 pattern at the low end. For example a low run of 3 bits
1111100111000000 run of bits
> one shifts up, rest drop to low end
1111101000001010
^ ^ ^
up low end
The final value in a 2*w block has all 1s at the high end. The first of the next bigger block of values is an alternating 1010..10. For example
111000 last 6bit value, all 1bits at high end
>
10101010 first 8bit value
Ith
As described above there are Catalan(w) many values with 2*w bits. The width of the i'th value can be found by successively subtracting C(1), C(2), etc until reaching a remainder i < C(w). At that point the value is 2*w many bits, being w many "1"s and w many "0"s.
In general after outputting some bits of the value (at the high end) there will be some number z many "0"s and n many "1"s yet to be output. The choice is then to output either 0 or 1 and reduce z or n accordingly.
numvalues(z,n) = number of sequences of z "0"s and n "1"s
with remaining 1s >= remaining 0s at all times
N = numvalues(z1,n)
= how many combinations starting with zero "0..."
output
0 if i < N
1 if i >= N
and subtract N from i
which is the "0..." combinations skipped
numvalues() is the "Catalan table" constructed by
for z=1 upwards
numvalues(z,0) = 1
for n = 1 to z
numvalues(z,n) = numvalues(z1,n) # the 0... forms
+ numvalues(z,n1) # the 1... forms
Each forming numvalues(z,n) the numvalues(z,n1) term is the previous numvalues calculated, so a simple addition loop for the table
for z=1 upwards
t = numvalues(z,0) = 1
for n = 1 to z
t += numvalues(z1,n)
numvalues(z,n) = t
The last entry numvalues(w,w) in each row is Catalan(w), so that can be used for the initial i subtractions seeking the width w. If building or extending a table each time then stop the table at that point.
Catalan(w) grows as a little less than a power 4^w so the table has a little more than log4(i) many rows.
SEE ALSO
Math::NumSeq, Math::NumSeq::Catalan, Math::NumSeq::Fibbinary
HOME PAGE
http://user42.tuxfamily.org/mathnumseq/index.html
LICENSE
Copyright 2012, 2013 Kevin Ryde
MathNumSeq is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 3, or (at your option) any later version.
MathNumSeq is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with MathNumSeq. If not, see <http://www.gnu.org/licenses/>.