Some Arithmetic, Algebraic andCombinatorial Aspects of Plane Binary Trees
Raazesh Sainudiin†
Part I with: Jennifer Harlow† & Warwick Tucker (Maths@Uppsala-SW)
Part II with: Sean Cleary (Maths@CCNY-USA), Robert Griffiths (Stats@Oxford-UK),
Mareike Fischer (Maths&CompSci@Greifswald-DL) & David Welch (CompSci@Auckland-NZ)
†School of Mathematics and Statistics, University of Canterbury,Christchurch, New Zealand
(On Sabbatical, Dept. of Mathematics, Cornell University, Ithaca, New York, USA
October 27 2014,Cornell Discrete Geometry & Combinatorics Seminar
1 / 94
Part I: Arithmetic and Algebra over Plane Binary TreesMain Idea & Motivating ExamplesRegular Pavings (RPs)Mapped Regular Pavings (MRPs)Real Mapped Regular Pavings (R-MRPs)Applications of Mapped Regular Pavings (MRPs)Conclusions of Part I
Part II: Combinatorics for Distributions over Plane Binary TreesCatalan CoefficientsSplit-Path Invariant DistributionsConclusions of Part II
2 / 94
Part I: Arithmetic and Algebra over Plane Binary Trees
3 / 94
Extending Arithmetic:reals→ intervals→ mapped partitions of interval
1. arithmetic over reals, eg. 1 + 3 = 4
2. naturally extends toarithmetic over intervals, eg. [1,2] + [3,4] = [4,6]
3. Our Main Idea:– is to further naturally extend toarithmetic over mapped partitions of an interval calledMapped Regular Pavings (MRPs)
4. – by exploiting the algebraic structure of partitions formedby rooted-plane-binary (rpb) trees
5. – thereby provide algorithms for several algebras and theirinclusions over rpb tree partitions
4 / 94
Extending Arithmetic:reals→ intervals→ mapped partitions of interval
1. arithmetic over reals, eg. 1 + 3 = 42. naturally extends to
arithmetic over intervals, eg. [1,2] + [3,4] = [4,6]
3. Our Main Idea:– is to further naturally extend toarithmetic over mapped partitions of an interval calledMapped Regular Pavings (MRPs)
4. – by exploiting the algebraic structure of partitions formedby rooted-plane-binary (rpb) trees
5. – thereby provide algorithms for several algebras and theirinclusions over rpb tree partitions
5 / 94
Extending Arithmetic:reals→ intervals→ mapped partitions of interval
1. arithmetic over reals, eg. 1 + 3 = 42. naturally extends to
arithmetic over intervals, eg. [1,2] + [3,4] = [4,6]3. Our Main Idea:
– is to further naturally extend toarithmetic over mapped partitions of an interval calledMapped Regular Pavings (MRPs)
4. – by exploiting the algebraic structure of partitions formedby rooted-plane-binary (rpb) trees
5. – thereby provide algorithms for several algebras and theirinclusions over rpb tree partitions
6 / 94
Extending Arithmetic:reals→ intervals→ mapped partitions of interval
1. arithmetic over reals, eg. 1 + 3 = 42. naturally extends to
arithmetic over intervals, eg. [1,2] + [3,4] = [4,6]3. Our Main Idea:
– is to further naturally extend toarithmetic over mapped partitions of an interval calledMapped Regular Pavings (MRPs)
4. – by exploiting the algebraic structure of partitions formedby rooted-plane-binary (rpb) trees
5. – thereby provide algorithms for several algebras and theirinclusions over rpb tree partitions
7 / 94
Extending Arithmetic:reals→ intervals→ mapped partitions of interval
1. arithmetic over reals, eg. 1 + 3 = 42. naturally extends to
arithmetic over intervals, eg. [1,2] + [3,4] = [4,6]3. Our Main Idea:
– is to further naturally extend toarithmetic over mapped partitions of an interval calledMapped Regular Pavings (MRPs)
4. – by exploiting the algebraic structure of partitions formedby rooted-plane-binary (rpb) trees
5. – thereby provide algorithms for several algebras and theirinclusions over rpb tree partitions
8 / 94
arithmetic from intervals to their rpb-tree partitions
Figure: Arithmetic with coloured spaces.
9 / 94
arithmetic from intervals to their rpb-tree partitions
Figure: Intersection of two hollow spheres.
10 / 94
arithmetic from intervals to their rpb-tree partitions
Figure: Histogram averaging.
11 / 94
An RP tree a root interval xρ ∈ IRd
The regularly paved boxes of xρ can be represented by nodes ofrooted-plane-binary (rpb) trees of enumerative combinatorics
finite-rooted-binary (frb) trees of geometric group theoryAn operation of bisection on a box is equivalent to performing the operation on its corresponding node in the tree:
Leaf boxes of RP tree partition the root interval xρ ∈ IR2
zρ
xρ
zρzρL
zρR
���
@@@
xρL xρR
zρ�
��z
���zρLL
@@@zρLR
@@@zρR
xρLR
xρLL
xρR
z�z
���zρLL
@@@zρLR
@@@zzρRL
@@@zρRR
���
xρLR
xρLL xρRL
xρRR
zρ�
��
@@@z
���
AAA
z���
AAAz
ρLL
z zρRL
zρRR
���
AAAz
ρLRL
zρLRR
xρLR
L
xρLR
R
xρLL xρRL
xρRR
By this “RP Peano’s curve” rpb-trees encode paritions of xρ ∈ IRd
12 / 94
An RP tree a root interval xρ ∈ IRd
The regularly paved boxes of xρ can be represented by nodes ofrooted-plane-binary (rpb) trees of enumerative combinatorics
finite-rooted-binary (frb) trees of geometric group theoryAn operation of bisection on a box is equivalent to performing the operation on its corresponding node in the tree:
Leaf boxes of RP tree partition the root interval xρ ∈ IR1
~ρ
xρ
~ρ
~ρL
~ρR
���
���
@@@@@@
xρL xρR
~ρ��
����~
��
����~
ρLL
@@@@@@~ρLR
@@@@@@~ρR
xρLRxρLL xρR
Leaf boxes of RP tree partition the root interval xρ ∈ IR2
zρ
xρ
zρzρL
zρR
���
@@@
xρL xρR
zρ�
��z
���zρLL
@@@zρLR
@@@zρR
xρLR
xρLL
xρR
z�z
���zρLL
@@@zρLR
@@@zzρRL
@@@zρRR
���
xρLR
xρLL xρRL
xρRR
zρ�
��
@@@z
���
AAA
z���
AAAz
ρLL
z zρRL
zρRR
���
AAAz
ρLRL
zρLRR
xρLR
L
xρLR
R
xρLL xρRL
xρRR
By this “RP Peano’s curve” rpb-trees encode paritions of xρ ∈ IRd
13 / 94
An RP tree a root interval xρ ∈ IRd
The regularly paved boxes of xρ can be represented by nodes ofrooted-plane-binary (rpb) trees of enumerative combinatorics
finite-rooted-binary (frb) trees of geometric group theoryAn operation of bisection on a box is equivalent to performing the operation on its corresponding node in the tree:
Leaf boxes of RP tree partition the root interval xρ ∈ IR2
zρ
xρ
zρzρL
zρR
���
@@@
xρL xρR
zρ�
��z
���zρLL
@@@zρLR
@@@zρR
xρLR
xρLL
xρR
z�z
���zρLL
@@@zρLR
@@@zzρRL
@@@zρRR
���
xρLR
xρLL xρRL
xρRR
zρ�
��
@@@z
���
AAA
z���
AAAz
ρLL
z zρRL
zρRR
���
AAAz
ρLRL
zρLRR
xρLR
L
xρLR
R
xρLL xρRL
xρRR
By this “RP Peano’s curve” rpb-trees encode paritions of xρ ∈ IRd
14 / 94
An RP tree a root interval xρ ∈ IRd
The regularly paved boxes of xρ can be represented by nodes ofrooted-plane-binary (rpb) trees of enumerative combinatorics
finite-rooted-binary (frb) trees of geometric group theoryAn operation of bisection on a box is equivalent to performing the operation on its corresponding node in the tree:
Leaf boxes of RP tree partition the root interval xρ ∈ IR2
zρ
xρ
zρzρL
zρR
���
@@@
xρL xρR
zρ�
��z
���zρLL
@@@zρLR
@@@zρR
xρLR
xρLL
xρR
z�z
���zρLL
@@@zρLR
@@@zzρRL
@@@zρRR
���
xρLR
xρLL xρRL
xρRR
zρ�
��
@@@z
���
AAA
z���
AAAz
ρLL
z zρRL
zρRR
���
AAAz
ρLRL
zρLRR
xρLR
L
xρLR
R
xρLL xρRL
xρRR
By this “RP Peano’s curve” rpb-trees encode paritions of xρ ∈ IRd15 / 94
Algebraic Structure and Combinatorics of RPs
Leaf-depth encoded RPs
There are Ck RPs with k splits
C0 = 1C1 = 1C2 = 2C3 = 5C4 = 14C5 = 42. . . = . . .
Ck =(2k)!
(k+1)!k!. . . = . . .C15 = 9694845. . . = . . .C20 = 6564120420. . . = . . .
16 / 94
Hasse (transition) Diagram of Regular Pavings
Transition diagram over S0:3 with split/reunion operations
RS, W.Taylor and G.Teng, Catalan Coefficients, Sequence A185155 in The On-Line Encyclopedia of Integer
Sequences, 2012, http://oeis.org
17 / 94
http://oeis.org
Hasse (transition) Diagram of Regular Pavings
Transition diagram over S0:4 with split/reunion operations
1. The above state space is denoted by S0:42. Number of RPs with k splits is the Catalan number Ck3. There is more than one way to reach a RP by k splits4. Randomized enclosure algorithms are Markov chains on
S0:∞
18 / 94
RPs are closed under union operations
s(1) ∪ s(2) = s is union of two RPs s(1) and s(2) of xρ ∈ R2.
s(1)
zρ(1)�
��
@@@zρ(1)L
���
@@@zρ(1)LL zρ(1)LR
zρ(1)R
s(2)
zρ(2)�
��
@@@zρ(2)L zρ(2)R�
��
@@@zρ(2)RL zρ(2)RR
s
zρ�
��
@@@zρL
���
@@@zρLL zρLR
zρR�
��@@@
zρRL
zρRR
xρ(1)LR
xρ(1)LL
xρ(1)R
∪
xρ(2)RR
xρ(2)RL
xρ(2)L
=
xρLR
xρLL xρRL
xρRR
Lemma 1: The algebraic structure of rpb-trees (underlyingThompson’s group) is closed under union operations.
Proof: by a “transparency overlay process” argument (cf. Meier2008).
s(1) ∪ s(2) = s is union of two RPs s(1) and s(2) of xρ ∈ R2.
19 / 94
RPs are closed under union operations
Lemma 1: The algebraic structure of rpb-trees (underlyingThompson’s group) is closed under union operations.
Proof: by a “transparency overlay process” argument (cf. Meier2008).
s(1) ∪ s(2) = s is union of two RPs s(1) and s(2) of xρ ∈ R2.
20 / 94
RPs are closed under union operations
Lemma 1: The algebraic structure of rpb-trees (underlyingThompson’s group) is closed under union operations.
Proof: by a “transparency overlay process” argument (cf. Meier2008).
s(1) ∪ s(2) = s is union of two RPs s(1) and s(2) of xρ ∈ R2.
21 / 94
Algorithm 1: RPUnion(ρ(1), ρ(2))input : Root nodes ρ(1) and ρ(2) of RPs s(1) and s(2) , respectively, with root box x
ρ(1)= x
ρ(2)
output : Root node ρ of RP s = s(1) ∪ s(2)
if IsLeaf(ρ(1)) & IsLeaf(ρ(2)) thenρ← Copy(ρ(1))return ρ
end
else if !IsLeaf(ρ(1)) & IsLeaf(ρ(2)) thenρ← Copy(ρ(1))return ρ
end
else if IsLeaf(ρ(1)) & !IsLeaf(ρ(2)) thenρ← Copy(ρ(2))return ρ
end
else!IsLeaf(ρ(1)) & !IsLeaf(ρ(2))
endMake ρ as a node with xρ ← xρ(1)Graft onto ρ as left child the node RPUnion(ρ(1)L, ρ(2)L)Graft onto ρ as right child the node RPUnion(ρ(1)R, ρ(2)R)return ρ
Note: this is not the minimal union of the (Boolean mapped) RPs of Jaulin et. al. 2001
22 / 94
Dfn: Mapped Regular Paving (MRP)
I Let s ∈ S0:∞ be an RP with root node ρ and root boxxρ ∈ IRd
I and let Y be a non-empty set.I Let V(s) and L(s) denote the sets all nodes and leaf nodes
of s, respectively.I Let f : V(s)→ Y map each node of s to an element in Y as
follows:{ρv 7→ fρv : ρv ∈ V(s), fρv ∈ Y} .
I Such a map f is called a Y-mapped regular paving(Y-MRP).
I Thus, a Y-MRP f is obtained by augmenting each node ρvof the RP tree s with an additional data member fρv.
23 / 94
Dfn: Mapped Regular Paving (MRP)
I Let s ∈ S0:∞ be an RP with root node ρ and root boxxρ ∈ IRd
I and let Y be a non-empty set.
I Let V(s) and L(s) denote the sets all nodes and leaf nodesof s, respectively.
I Let f : V(s)→ Y map each node of s to an element in Y asfollows:
{ρv 7→ fρv : ρv ∈ V(s), fρv ∈ Y} .
I Such a map f is called a Y-mapped regular paving(Y-MRP).
I Thus, a Y-MRP f is obtained by augmenting each node ρvof the RP tree s with an additional data member fρv.
24 / 94
Dfn: Mapped Regular Paving (MRP)
I Let s ∈ S0:∞ be an RP with root node ρ and root boxxρ ∈ IRd
I and let Y be a non-empty set.I Let V(s) and L(s) denote the sets all nodes and leaf nodes
of s, respectively.
I Let f : V(s)→ Y map each node of s to an element in Y asfollows:
{ρv 7→ fρv : ρv ∈ V(s), fρv ∈ Y} .
I Such a map f is called a Y-mapped regular paving(Y-MRP).
I Thus, a Y-MRP f is obtained by augmenting each node ρvof the RP tree s with an additional data member fρv.
25 / 94
Dfn: Mapped Regular Paving (MRP)
I Let s ∈ S0:∞ be an RP with root node ρ and root boxxρ ∈ IRd
I and let Y be a non-empty set.I Let V(s) and L(s) denote the sets all nodes and leaf nodes
of s, respectively.I Let f : V(s)→ Y map each node of s to an element in Y as
follows:{ρv 7→ fρv : ρv ∈ V(s), fρv ∈ Y} .
I Such a map f is called a Y-mapped regular paving(Y-MRP).
I Thus, a Y-MRP f is obtained by augmenting each node ρvof the RP tree s with an additional data member fρv.
26 / 94
Dfn: Mapped Regular Paving (MRP)
I Let s ∈ S0:∞ be an RP with root node ρ and root boxxρ ∈ IRd
I and let Y be a non-empty set.I Let V(s) and L(s) denote the sets all nodes and leaf nodes
of s, respectively.I Let f : V(s)→ Y map each node of s to an element in Y as
follows:{ρv 7→ fρv : ρv ∈ V(s), fρv ∈ Y} .
I Such a map f is called a Y-mapped regular paving(Y-MRP).
I Thus, a Y-MRP f is obtained by augmenting each node ρvof the RP tree s with an additional data member fρv.
27 / 94
Dfn: Mapped Regular Paving (MRP)
I Let s ∈ S0:∞ be an RP with root node ρ and root boxxρ ∈ IRd
I and let Y be a non-empty set.I Let V(s) and L(s) denote the sets all nodes and leaf nodes
of s, respectively.I Let f : V(s)→ Y map each node of s to an element in Y as
follows:{ρv 7→ fρv : ρv ∈ V(s), fρv ∈ Y} .
I Such a map f is called a Y-mapped regular paving(Y-MRP).
I Thus, a Y-MRP f is obtained by augmenting each node ρvof the RP tree s with an additional data member fρv.
28 / 94
Examples of Y-MRPs
If Y = R
R-MRP over s221 with xρ = [0,8]
29 / 94
Examples of Y-MRPs
If Y = B
B-MRP over s122 with xρ = [0,1]2 (e.g. Jaulin et. al. 2001)
30 / 94
Examples of Y-MRPs
If Y = IR– rpb tree representation for interval inclusion algebra
IR-MRP enclosure of the Rosenbrock function withxρ = [−1,1]2
31 / 94
Examples of Y-MRPs
If Y = [0,1]3– R G B colour maps
[0,1]3-MRP over s3321 with xρ = [0,1]3
32 / 94
Examples of Y-MRPs
If Y = Z+ := {0,1,2, ...}– radar-measured aircraft trajectory data
Z+-MRP trajectory of an aircraft and its tree
33 / 94
Examples of Y-MRPs
If Y = S2, xρ = [0,1]2 – vector-MRPs
Two Views of f
Two Views of g
34 / 94
Y-MRP Arithmetic
If ? : Y× Y→ Y then we can extend ? point-wise to twoY-MRPs f and g with root nodes ρ(1) and ρ(2) viaMRPOperate(ρ(1), ρ(2), ?).This is done using MRPOperate(ρ(1), ρ(2),+)
f g f + g
35 / 94
R-MRP Addition by MRPOperate(ρ(1), ρ(2),+)
adding two piece-wise constant functions or R-MRPs
36 / 94
Algorithm 2: MRPOperate(ρ(1), ρ(2), ?)input : two root nodes ρ(1) and ρ(2) with same root box x
ρ(1)= x
ρ(2)and binary operation ?.
output : the root node ρ of Y-MRP h = f ? g.Make a new node ρ with box and imagexρ ← xρ(1) ; hρ ← fρ(1) ? gρ(2)
if IsLeaf(ρ(1)) & !IsLeaf(ρ(2)) thenMake temporary nodes L′, R′
xL′ ← xρ(1)L; xR′ ← xρ(1)RfL′ ← fρ(1) , fR′ ← fρ(1)Graft onto ρ as left child the node MRPOperate(L′, ρ(2)L, ?)Graft onto ρ as right child the node MRPOperate(R′, ρ(2)R, ?)
end
else if !IsLeaf(ρ(1)) & IsLeaf(ρ(2)) thenMake temporary nodes L′, R′
xL′ ← xρ(2)L; xR′ ← xρ(2)RgL′ ← gρ(2) , gR′ ← gρ(2)Graft onto ρ as left child the node MRPOperate(ρ(1)L, L′, ?)Graft onto ρ as right child the node MRPOperate(ρ(1)R,R′, ?)
end
else if !IsLeaf(ρ(1)) & !IsLeaf(ρ(2)) thenGraft onto ρ as left child the node MRPOperate(ρ(1)L, ρ(2)L, ?)Graft onto ρ as right child the node MRPOperate(ρ(1)R, ρ(2)R, ?)
endreturn ρ
37 / 94
Unary transformations are easy too
Let MRPTransform(ρ, τ) apply the unary transformationτ : R→ R to a given R-MRP f with root node ρ as follows:
I copy f to gI recursively set fρv = τ(fρv ) for each node ρv in gI return g as τ(f )
38 / 94
Minimal Representation of R-MRP
Algorithm 3: MinimiseLeaves(ρ)input : ρ, the root node of R-MRP f .output : Modify f into h(f ), the unique R-MRP with fewest leaves.if !IsLeaf(ρ) then
MinimiseLeaves(ρL)MinimiseLeaves(ρR)
if IsCherry(ρ) & ( fρL = fρR ) thenfρ ← fρLPrune(ρL)Prune(ρR)
endend
39 / 94
Arithmetic and Algebra of R-MRPs
Thus, we can obtain arithmetical expressions specified byfinitely many sub-expressions in a directed acyclic graphwhose:
I inputs and output nodes are themselves R-MRPsI and whose edges involve:
1. a binary arithmetic operation ? ∈ {+,−, ·, /} over twoR-MRPs,
2. a standard transformation of R-MRP by elements ofS := {exp, sin, cos, tan, . . .} and
3. their compositions.
40 / 94
Arithmetic and Algebra of R-MRPs
Thus, we can obtain arithmetical expressions specified byfinitely many sub-expressions in a directed acyclic graphwhose:
I inputs and output nodes are themselves R-MRPs
I and whose edges involve:1. a binary arithmetic operation ? ∈ {+,−, ·, /} over two
R-MRPs,2. a standard transformation of R-MRP by elements of
S := {exp, sin, cos, tan, . . .} and
3. their compositions.
41 / 94
Arithmetic and Algebra of R-MRPs
Thus, we can obtain arithmetical expressions specified byfinitely many sub-expressions in a directed acyclic graphwhose:
I inputs and output nodes are themselves R-MRPsI and whose edges involve:
1. a binary arithmetic operation ? ∈ {+,−, ·, /} over twoR-MRPs,
2. a standard transformation of R-MRP by elements ofS := {exp, sin, cos, tan, . . .} and
3. their compositions.
42 / 94
Arithmetic and Algebra of R-MRPs
Thus, we can obtain arithmetical expressions specified byfinitely many sub-expressions in a directed acyclic graphwhose:
I inputs and output nodes are themselves R-MRPsI and whose edges involve:
1. a binary arithmetic operation ? ∈ {+,−, ·, /} over twoR-MRPs,
2. a standard transformation of R-MRP by elements ofS := {exp, sin, cos, tan, . . .} and
3. their compositions.
43 / 94
Stone-Weierstrass Theorem: R-MRPs Dense in C(xρ,R)
TheoremLet F be the class of R-MRPs with the same root box xρ. ThenF is dense in C(xρ,R), the algebra of real-valued continuousfunctions on xρ.
Proof:Since xρ ∈ IRd is a compact Hausdorff space, by theStone-Weierstrass theorem we can establish that F is dense inC(xρ,R) with the topology of uniform convergence, providedthat F is a sub-algebra of C(xρ,R) that separates points in xρand which contains a non-zero constant function.
We will show all these conditions are satisfied by F
44 / 94
Stone-Weierstrass Theorem: R-MRPs Dense in C(xρ,R)
TheoremLet F be the class of R-MRPs with the same root box xρ. ThenF is dense in C(xρ,R), the algebra of real-valued continuousfunctions on xρ.Proof:Since xρ ∈ IRd is a compact Hausdorff space, by theStone-Weierstrass theorem we can establish that F is dense inC(xρ,R) with the topology of uniform convergence, providedthat F is a sub-algebra of C(xρ,R) that separates points in xρand which contains a non-zero constant function.
We will show all these conditions are satisfied by F
45 / 94
Stone-Weierstrass Theorem: R-MRPs Dense in C(xρ,R)
TheoremLet F be the class of R-MRPs with the same root box xρ. ThenF is dense in C(xρ,R), the algebra of real-valued continuousfunctions on xρ.Proof:Since xρ ∈ IRd is a compact Hausdorff space, by theStone-Weierstrass theorem we can establish that F is dense inC(xρ,R) with the topology of uniform convergence, providedthat F is a sub-algebra of C(xρ,R) that separates points in xρand which contains a non-zero constant function.
We will show all these conditions are satisfied by F
46 / 94
Stone-Weierstrass Theorem Contd.: R-MRPs Dense inC(xρ,R)
I F is a sub-algebra of C(xρ,R) since it is closed underaddition and scalar multiplication.
I F contains non-zero constant functionsI Finally, RPs can clearly separate distinct points x , x ′ ∈ xρ
into distinct leaf boxes by splitting deeply enough.I Thus, F , the class of R-MRPs with the same root box xρ,
is dense in C(xρ,R), the algebra of real-valued continuousfunctions on xρ.
I Q.E.D.
47 / 94
Stone-Weierstrass Theorem Contd.: R-MRPs Dense inC(xρ,R)
I F is a sub-algebra of C(xρ,R) since it is closed underaddition and scalar multiplication.
I F contains non-zero constant functions
I Finally, RPs can clearly separate distinct points x , x ′ ∈ xρinto distinct leaf boxes by splitting deeply enough.
I Thus, F , the class of R-MRPs with the same root box xρ,is dense in C(xρ,R), the algebra of real-valued continuousfunctions on xρ.
I Q.E.D.
48 / 94
Stone-Weierstrass Theorem Contd.: R-MRPs Dense inC(xρ,R)
I F is a sub-algebra of C(xρ,R) since it is closed underaddition and scalar multiplication.
I F contains non-zero constant functionsI Finally, RPs can clearly separate distinct points x , x ′ ∈ xρ
into distinct leaf boxes by splitting deeply enough.
I Thus, F , the class of R-MRPs with the same root box xρ,is dense in C(xρ,R), the algebra of real-valued continuousfunctions on xρ.
I Q.E.D.
49 / 94
Stone-Weierstrass Theorem Contd.: R-MRPs Dense inC(xρ,R)
I F is a sub-algebra of C(xρ,R) since it is closed underaddition and scalar multiplication.
I F contains non-zero constant functionsI Finally, RPs can clearly separate distinct points x , x ′ ∈ xρ
into distinct leaf boxes by splitting deeply enough.I Thus, F , the class of R-MRPs with the same root box xρ,
is dense in C(xρ,R), the algebra of real-valued continuousfunctions on xρ.
I Q.E.D.
50 / 94
Stone-Weierstrass Theorem Contd.: R-MRPs Dense inC(xρ,R)
I F is a sub-algebra of C(xρ,R) since it is closed underaddition and scalar multiplication.
I F contains non-zero constant functionsI Finally, RPs can clearly separate distinct points x , x ′ ∈ xρ
into distinct leaf boxes by splitting deeply enough.I Thus, F , the class of R-MRPs with the same root box xρ,
is dense in C(xρ,R), the algebra of real-valued continuousfunctions on xρ.
I Q.E.D.
51 / 94
Approximating Kernel Density Estimates by R-MRPs
52 / 94
Approximating Kernel Density Estimates by R-MRPs
53 / 94
Approximating Kernel Density Estimates by R-MRPs
54 / 94
Nonparametric Density Estimation
Problem: Take samples from an unknown density f and consistentlyreconstruct f
55 / 94
Nonparametric Density Estimation
Approach: Use statistical regular paving to get R-MRP data-adaptivehistogram
56 / 94
Nonparametric Density Estimation
Solution: R-MRP histogram averaging allows us to produce aconsistent Bayesian estimate of the density (up to 10 dimensions)(Teng, Harlow, Lee and S., ACM Trans. Mod. & Comp. Sim., [r. 2] 2012)
57 / 94
Coverage, Marginal & Slice Operators of R-MRP
R-MRP approximation to Levy density and its coverage regions withα = 0.9 (light gray), α = 0.5 (dark gray) and α = 0.1 (black)
58 / 94
Coverage, Marginal & Slice Operators of R-MRP
Marginal densities f {1}(x1) and f {2}(x2) along each coordinate ofR-MRP approximation
59 / 94
Coverage, Marginal & Slice Operators of R-MRP
The slices of a simple R-MRP in 2D
— “non-parametric regression arithmetic”
60 / 94
[0,1]3-MRP Arithmetic over Colored Cubes
f g
f + g f − g
61 / 94
Slices of colored cube
62 / 94
B-MRP arithmetic – contractors, propagators &collaborators (Comp-aided Proofs in Anal./Dyn.)
Two Boolean-mapped regular pavings A1 and A2 and Booleanarithmetic operations with + for set union, − for symmetric set
difference, × for set intersection, and ÷ for set difference.
A1 A2 A1 + A2
63 / 94
B-MRP arithmetic – contractors, propagators &collaborators (Comp-aided Proofs in Anal./Dyn.)
Two Boolean-mapped regular pavings A1 and A2 and Booleanarithmetic operations with + for set union, − for symmetric set
difference, × for set intersection, and ÷ for set difference.
A1 − A2 A1 × A2 A1 ÷ A2
64 / 94
Vector-MRP arithmetic
If Y = S2, xρ = [0,1]2 – vector-MRPs
Two Views of f
Two Views of g
65 / 94
Vector-MRP arithmetic
f × g — cross-product of vector-MRPs
66 / 94
Air Traffic “Arithmetic”
(G. Teng, K. Kuhn and RS, J. Aerospace Comput., Inf. & Com., 9:1, 14–25, 2012.)
On a Good Day
67 / 94
Air Traffic “Arithmetic”
(G. Teng, K. Kuhn and RS, J. Aerospace Comput., Inf. & Com., 9:1, 14–25, 2012.)
Z+-MRP On a Good Day
68 / 94
Air Traffic “Arithmetic”
(G. Teng, K. Kuhn and RS, J. Aerospace Comput., Inf. & Com., 9:1, 14–25, 2012.)
On a Bad Day
69 / 94
Air Traffic “Arithmetic”
(G. Teng, K. Kuhn and RS, J. Aerospace Comput., Inf. & Com., 9:1, 14–25, 2012.)
Z+-MRP On a Bad Day
70 / 94
Air Traffic “Arithmetic”
(G. Teng, K. Kuhn and RS, J. Aerospace Comput., Inf. & Com., 9:1, 14–25, 2012.)
Z-MRP pattern for Good Day − Bad Day
71 / 94
MRS 1.0: A C++ Class Library for Statistical Set Processing,
Harlow, S & York, 2013
MRS 1.0 is GNU auto-confiscated, Doxygenized, GPL-licensed (builds on GNU Sci. Lib., C-XSC & Boost) and has:
Its is templatised and can be extended to general Y-MRPs.
72 / 94
Conclusions of Part I
I Y-MRPs provide rpb-tree partition arithmeticI IY-MRPs allow efficient arithmetic for Neumaier’s inclusion
algebrasI I IY can be IR for f : IRd → IR
I IY can be IRm for f : IRd → IRmI IY can be (IR, IRm, IRm2) for range, gradient & Hessian of
f : IRd → IRI Other obvious extensions include arithmetic over Taylor
polynomial inclusion algebrasI In general the domain and range of f can be complete
lattices with intervals and bisection operationsI We have seen several statistical applications of Y-MRPsI CODE: mrs: a C++ class library for statistical set
processing by Harlow, S and York.
73 / 94
Part II: Combinatorics for Distributions over Plane Binary Trees
74 / 94
Catalan Coefficients
How many distinct “splitting paths” are there from the root nodeto a given rpb-tree T ?Let this be B(T ), the Catalan Coefficient of T .
1 1 2 1 1
There are Ck RPs with k splits
and k! distinct paths to them
k Ck k!0 1 11 1 12 2 23 5 64 14 245 42 120. . . . . . . . .
k (2k)!(k+1)!k! k!
. . . . . . . . .
1 1 2 1 1
For example: B((2,2,2,2)) = 2 and B((3,3,2,1)) = 1
75 / 94
Catalan coefficients – OEIS A185155
76 / 94
Catalan coefficients of rpb-trees with 3,4,5 splits
77 / 94
Frequency of Catalan coefficients of rpb-trees with6,7,8 splits
78 / 94
Catalan coefficient of a rpb-tree
Let an interior node of T include the root node and exclude allleaf nodes of T . Then the Catalan coefficient of T is:
B(T ) =|T |!∏
v∈T|Tv |
=(# of interior nodes of T )!∏
v∈T# of interior nodes of sub-tree of T with root v
Proof:
79 / 94
Catalan coefficient of a rpb-tree
Let an interior node of T include the root node and exclude allleaf nodes of T . Then the Catalan coefficient of T is:
B(T ) =|T |!∏
v∈T|Tv |
=(# of interior nodes of T )!∏
v∈T# of interior nodes of sub-tree of T with root v
Proof:Let L(Tv ) and R(Tv ) be left and right sub-trees of Tv with rootnode v . Then the number of distinct binary inter-leavingsbetween the interior (split) nodes of L(Tv ) and R(Tv ) is:(
|L(Tv )|+ |R(Tv )||L(Tv )|
)=
(|L(Tv )|+ |R(Tv )|)!|L(Tv )|!× |R(Tv )|!
=|Tv | × (|L(Tv )|+ |R(Tv )|)!|Tv | × |L(Tv )|!× |R(Tv )|!
=|Tv |!
|Tv | × |L(Tv )|!× |R(Tv )|!
80 / 94
Catalan coefficient of a rpb-tree
Let an interior node of T include the root node and exclude allleaf nodes of T . Then the Catalan coefficient of T is:
B(T ) =|T |!∏
v∈T|Tv |
=(# of interior nodes of T )!∏
v∈T# of interior nodes of sub-tree of T with root v
Proof:And the number of distinct binary inter-leavings between theinterior (split) nodes of L(Tv ) and R(Tv ) as well as theirsub-trees and their sub-sub-trees and so on is:
|Tv |!|Tv | ����|L(Tv )|!���
��|R(Tv )|!×
����|L(Tv )|!
|L(Tv )| × |L(L(Tv ))|!× |R(L(Tv ))|!×
�����|R(Tv )|!
|R(Tv )| × |L(R(Tv ))|!× |R(R(Tv ))|!× · · · 1!
1× 0!× 0!
81 / 94
Catalan coefficient of a rpb-tree
Let an interior node of T include the root node and exclude allleaf nodes of T . Then the Catalan coefficient of T is:
B(T ) =|T |!∏
v∈T|Tv |
=(# of interior nodes of T )!∏
v∈T# of interior nodes of sub-tree of T with root v
Proof:And the number of distinct binary inter-leavings between theinterior (split) nodes of L(Tv ) and R(Tv ) as well as theirsub-trees and their sub-sub-trees and so on is:
|Tv |!|Tv | ����|L(Tv )|!���
��|R(Tv )|!×
����|L(Tv )|!
|L(Tv )| ������|L(L(Tv ))|!���
���|R(L(Tv ))|!×
�����|R(Tv )|!
|R(Tv )| ×������|L(R(Tv ))|!×(((((
(|R(R(Tv ))|!× · · · ��1!
1��0!��0!
82 / 94
Catalan coefficient of a rpb-tree
Let an interior node of T include the root node and exclude allleaf nodes of T . Then the Catalan coefficient of T is:
B(T ) =|T |!∏
v∈T|Tv |
=(# of interior nodes of T )!∏
v∈T# of interior nodes of sub-tree of T with root v
Proof:
B(Tv ) =|Tv |!
|Tv | × |L(Tv )|!× |R(Tv )|!× B(L(Tv ))× B(R(Tv ))
=|Tv |!
|Tv | × |L(Tv )| × |R(Tv )| × |L(L(Tv ))| × |R(L(Tv ))| × · · ·
=|Tv |!∏
u∈Tv|Tu|
Therefore,
B(T ) =|T |!∏
v∈T |Tv |83 / 94
An Example Catalan coefficient calculation
Consider the perfectly balanced tree (3,3,3,3,3,3,3,3) withk = 7 splits and 8 leaves (all with depth 3).
Then B((3,3,3,3,3,3,3,3))
=7!
7× 3× 3× 1× 1× 1× 1=
2��6× 5× 4×��3× 2
��3��3= 80.
84 / 94
Planar Yule Model
Model (Planar Yule)A rpb-tree with k splits is obtained from one with k − 1 splits bysplitting one of the k leaves uniformly at random.
Pr(T ) = B(T )1k !
This model also gives:I the probability of a binary search tree with random input
given by the unifrom distribution on all permutations of [k ](Fill, 1994)
I the probability of an planar Yule tree (non-planar case isthe Yule speciation model in Phylogenetics)
85 / 94
Planar Yule Model
Model (Planar Yule)A rpb-tree with k splits is obtained from one with k − 1 splits bysplitting one of the k leaves uniformly at random.
Pr(T ) = B(T )1k !
This model also gives:I the probability of a binary search tree with random input
given by the unifrom distribution on all permutations of [k ](Fill, 1994)
I the probability of an planar Yule tree (non-planar case isthe Yule speciation model in Phylogenetics)
86 / 94
f -Splitting Model
Model (f -Splitting)Let f be a probability density function on [0,1]. We canrepresent the rpb-tree by the corresponding dyadic partition of[0,1]. We split a node v of an rpb-tree corresponding to theinterval [v , v ] = [2−i ,2−j ] with probability
∫ vv f (x)dx.
Pr(T ) = B(T )∏v∈T
(∫ vv
f (x)dx
)
This model also gives:I the Statistically equivalent block rule in density estimationI gives a flexible way to explore the tree shapes,
eg. f ∼ Beta(α, β)
87 / 94
f -Splitting Model
Model (f -Splitting)Let f be a probability density function on [0,1]. We canrepresent the rpb-tree by the corresponding dyadic partition of[0,1]. We split a node v of an rpb-tree corresponding to theinterval [v , v ] = [2−i ,2−j ] with probability
∫ vv f (x)dx.
Pr(T ) = B(T )∏v∈T
(∫ vv
f (x)dx
)
This model also gives:I the Statistically equivalent block rule in density estimationI gives a flexible way to explore the tree shapes,
eg. f ∼ Beta(α, β)
88 / 94
Split-Path Invariant Distributions
Model (Split-Path Invariant)
Pr(T ) = B(T )× Pr{any split-path to T from root node}
The Planar Yule Model and f -Splitting Model are examples ofSplit-Path Invariant Distributions. Such distributions onrpb-trees can also be used to model:
I infection trees in epidemics on various hidden contactgraphs (eq. all-to-all, one-super-hub, etc.)
I causal trees underlying self-exciting point processesI by counting planar tree paths that map to non-palanar tree
paths we can induce these distributions on non-planartrees
89 / 94
Split-Path Invariant Distributions
Model (Split-Path Invariant)
Pr(T ) = B(T )× Pr{any split-path to T from root node}
The Planar Yule Model and f -Splitting Model are examples ofSplit-Path Invariant Distributions. Such distributions onrpb-trees can also be used to model:
I infection trees in epidemics on various hidden contactgraphs (eq. all-to-all, one-super-hub, etc.)
I causal trees underlying self-exciting point processesI by counting planar tree paths that map to non-palanar tree
paths we can induce these distributions on non-planartrees
90 / 94
Conclusions of Part II
I Several probablistic models on rooted planar binary treesexist
I Using the notion of split-path invariance and Catalancoefficients we can specify various useful distributions onrpb-trees
I These distributions can also be further projected ontonon-planar binary trees
91 / 94
References
Jaulin, L., Kieffer, M., Didrit, O. & Walter, E., Applied intervalanalysis. London: Springer-Verlag, 2001.Meier, J., Groups, graphs and trees: an introduction to thegeometry of infinite groups, CUP, Cambridge, 2008.Neumaier, A., Interval methods for systems of equations, CUP,Cambridge, 1990.Harlow, J., Sainudiin, R. & Tucker, W., Mapped RegularPavings, Reliable Computing, vol. 16, pp. 252-282, 2012.Fill, J.A., On the stationary distribution for the move-to-rootMarkov chain for binary search trees, Tech. Rep. 537,Dept. Math.Sci., Johns Hopkins Univ., 1994.Sainudiin, R., Taylor, W., & Teng, G., Catalan Coefficients,Sequence A185155 in The On-Line Encyclopedia of IntegerSequences, published electronically at http://oeis.org,2012
92 / 94
http://oeis.org
Acknowledgements
I RS’s external consulting revenues from the New ZealandMinistry of Tourism
I WT’s Swedish Research Council Grant 2008-7510 thatenabled RS’s visits to Uppsala in 2006 and 2009
I Erskine grant from University of Canterbury that enabledWT’s visit to Christchurch in 2011 & 2014
I EU Marie Curie International Research Staff ExchangeGrants (CORCON 2014-2017 & CONSTRUMATH2009-2012)
I University of Canterbury MSc Scholarship to JH.
93 / 94
Thank you!
94 / 94
Part I: Arithmetic and Algebra over Plane Binary TreesMain Idea & Motivating ExamplesRegular Pavings (RPs)Mapped Regular Pavings (MRPs)Real Mapped Regular Pavings (R-MRPs)Applications of Mapped Regular Pavings (MRPs)Conclusions of Part I
Part II: Combinatorics for Distributions over Plane Binary TreesCatalan CoefficientsSplit-Path Invariant DistributionsConclusions of Part II
0.0: 0.1: 0.2: 0.3: 0.4: 0.5: 0.6: 0.7: 0.8: 0.9: anm0: