5. persistence diagrams...5. persistence diagrams vin de silva pomona college cbms lecture series,...
TRANSCRIPT
5. Persistence Diagrams
Vin de SilvaPomona College
CBMS Lecture Series, Macalester College, June 2017
Vin de Silva Pomona College 5. Persistence Diagrams
Persistence modules
Persistence modules
A diagram of vector spaces and linear maps
V0// V1
// . . . // Vn
is called a persistence module.
Vin de Silva Pomona College 5. Persistence Diagrams
Persistence modules
Persistence modules
A diagram of vector spaces and linear maps
V0// V1
// . . . // Vn
is called a persistence module.
Example
Let
S0⊆// S1
⊆// . . .
⊆// Sn
be a nested sequence of simplicial complexes. Then
Hk(S0) // Hk(S1) // . . . // Hk(Sn)
is a persistence module. The arrows denote the maps in homology induced bythe inclusions Sk−1 ⊆ Sk .
Vin de Silva Pomona College 5. Persistence Diagrams
Persistence modules
Persistence modules
A diagram of vector spaces and linear maps
V0// V1
// . . . // Vn
is called a persistence module.
Example
Let X be a finite metric space and let
r0 < r1 < · · · < rn
be the set of values taken by the metric. Then
Hk(VR(X , r0)) // Hk(VR(X , r1)) // . . . // Hk(VR(X , rn))
is a persistence module. The arrows denote the maps in homology induced bythe inclusions VR(X , rk−1) ⊆ VR(X , rk).
Vin de Silva Pomona College 5. Persistence Diagrams
Persistence modules
Persistence modules
A diagram of vector spaces and linear maps
V0// V1
// . . . // Vn
is called a persistence module.
Example
Let M be a compact smooth manifold and let f : M → R be a smooth functionwith finitely many critical values
t0 < t1 < · · · < tn.
Let M t = f −1(−∞, t]. Then
Hk(M t0 ) // Hk(M t1 ) // . . . // Hk(M tn )
is a persistence module. The arrows denote the maps in homology induced bythe inclusions M tk−1 ⊆ M tk .
Vin de Silva Pomona College 5. Persistence Diagrams
Persistence modules
Persistence modules
A persistence module indexed by N is an infinite diagram
V0// V1
// . . . // Vn// . . .
of vector spaces and linear maps.
Zomorodian, Carlsson (2002)
This is the same thing as a graded module over the polynomial ring F[t].
Field elements act on each Vk by scalar multiplication. The ring element t actsby applying the map Vk → Vk+1.
Vin de Silva Pomona College 5. Persistence Diagrams
The rank invariant
n=0
Diagrams of the formV0
are classified up to isomorphism by a single numerical invariant:
dim(V0)
Vin de Silva Pomona College 5. Persistence Diagrams
The rank invariant
n=0
Diagrams of the formV0
are classified up to isomorphism by a single numerical invariant:
r(0, 0) = rank(V0 → V0) = dim(V0)
Vin de Silva Pomona College 5. Persistence Diagrams
The rank invariant
n=1
Diagrams of the form
V0// V1
are classified up to isomorphism by three numerical invariants:
dim(V0)
rank(V0 → V1)
dim(V1)
Proof
Represent the map by a matrix. Changes of basis of V0,V1 correspond to columnand row operations. Every matrix can be reduced to the block form:[
Ir 00 0
]These block forms are precisely classified by the three invariants.
Vin de Silva Pomona College 5. Persistence Diagrams
The rank invariant
n=1
Diagrams of the form
V0// V1
are classified up to isomorphism by three numerical invariants:
r(0, 0) = dim(V0) = rank(V0 → V0)
r(0, 1) = rank(V0 → V1)
r(1, 1) = dim(V1) = rank(V1 → V1)
Proof
Represent the map by a matrix. Changes of basis of V0,V1 correspond to columnand row operations. Every matrix can be reduced to the block form:[
Ir 00 0
]These block forms are precisely classified by the three invariants.
Vin de Silva Pomona College 5. Persistence Diagrams
The rank invariant
Theorem (Zomorodian, Carlsson, 2002)
Diagrams of the form
V0// V1
// . . . // Vn
are classified up to isomorphism by the invariants
r(i , j) = rank(Vi → Vj)
for all 0 ≤ i ≤ j ≤ n.
Plot these as points-with-multiplicity on a ‘rank diagram’:
• • • • • 4
• • • • 3
• • • 2
• • 1
• 0
0 1 2 3 4 r(i, j)
Vin de Silva Pomona College 5. Persistence Diagrams
The rank invariant
Theorem (Zomorodian, Carlsson, 2002)
Diagrams of the form
V0// V1
// . . . // Vn
are classified up to isomorphism by the invariants
r(i , j) = rank(Vi → Vj)
for all 0 ≤ i ≤ j ≤ n.
Plot these as points-with-multiplicity on a ‘rank diagram’:
• • • • • 4
• • • • 3
• • • 2
• • 1
• 0
0 1 2 3 4 r(i, j)
Vin de Silva Pomona College 5. Persistence Diagrams
Interval decomposition
Direct sums
Given two persistence modules
V0// V1
// . . . // Vn
W0// W1
// . . . // Wn
we can form their direct sum:
V0 ⊕W0// V1 ⊕W1
// . . . // Vn ⊕Wn
Direct sum decomposition
Given a persistence module
V0// V1
// . . . // Vn
can it be expressed as the direct sum of two non-trivial persistence modules? Ifit can’t, then it is indecomposable.
Vin de Silva Pomona College 5. Persistence Diagrams
Interval decomposition
Interval modules
The interval module I (i , j) is defined to be the persistence module with
Vk =
{F if i ≤ k ≤ j
0 otherwise
and all maps F→ F being the identity (the other maps necessarily being zero).
Here are examples of interval modules with n = 5:
0 // 0 // F 1 // F 1 // F // 0
0 // 0 // F 1 // F 1 // F // F
F 1 // F 1 // F 1 // F 1 // F // F
Fact
Interval modules are indecomposable.
Vin de Silva Pomona College 5. Persistence Diagrams
Interval decomposition
Interval modules
The interval module I (i , j) is defined to be the persistence module with
Vk =
{F if i ≤ k ≤ j
0 otherwise
and all maps F→ F being the identity (the other maps necessarily being zero).
Here are examples of interval modules with n = 5:
0 // 0 // F 1 // F 1 // F // 0
0 // 0 // F 1 // F 1 // F // F
F 1 // F 1 // F 1 // F 1 // F // F
Fact
Interval modules are indecomposable.
Vin de Silva Pomona College 5. Persistence Diagrams
Interval decomposition
Theorem (Gabriel 1970; Zomorodian & Carlsson 2002; etc.)
Every (finite-dimensional) persistence module is isomorphic to a (finite) directsum of interval modules. The multiplicity
m(i , j)
of the summand I (i , j) is an invariant of the persistence module.
Proof
Zomorodian & Carlsson use the classification theorem for finitely-generatedgraded modules over F[t].
Plot the multiplicities as points-with-multiplicity on a ‘persistence diagram’:
• • • • • 4
• • • • 3
• • • 2
• • 1
• 0
0 1 2 3 4 m(i, j)
Vin de Silva Pomona College 5. Persistence Diagrams
Interval decomposition
Theorem (Gabriel 1970; Zomorodian & Carlsson 2002; etc.)
Every (finite-dimensional) persistence module is isomorphic to a (finite) directsum of interval modules. The multiplicity
m(i , j)
of the summand I (i , j) is an invariant of the persistence module.
Proof
Zomorodian & Carlsson use the classification theorem for finitely-generatedgraded modules over F[t].
Plot the multiplicities as points-with-multiplicity on a ‘persistence diagram’:
• • • • • 4
• • • • 3
• • • 2
• • 1
• 0
0 1 2 3 4 m(i, j)
Vin de Silva Pomona College 5. Persistence Diagrams
Interval decomposition
Theorem (Gabriel 1970; Zomorodian & Carlsson 2002; etc.)
Every (finite-dimensional) persistence module is isomorphic to a (finite) directsum of interval modules. The multiplicity
m(i , j)
of the summand I (i , j) is an invariant of the persistence module.
Proof
Zomorodian & Carlsson use the classification theorem for finitely-generatedgraded modules over F[t].
Plot the multiplicities as points-with-multiplicity on a ‘persistence diagram’:
• • • • • 4
• • • • 3
• • • 2
• • 1
• 0
0 1 2 3 4 m(i, j)
Vin de Silva Pomona College 5. Persistence Diagrams
Relationship between the two diagrams
Inverse transformations
• • • • • 4
• • • • 3
• • • 2
• • 1
• 0
0 1 2 3 4 r(i, j)
//oo
• • • • • 4
• • • • 3
• • • 2
• • 1
• 0
0 1 2 3 4 m(i, j)
rank diagram ← persistence diagram
Each summand I (p, q) contributes to r(i , j) whenever p ≤ i ≤ j ≤ q. Thus:
r(i , j) =∑i
p=0
∑nq=j m(p, q)
rank diagram → persistence diagram
The inverse transformation is
m(i , j) = r(i , j)− r(i − 1, j)− r(i , j + 1) + r(i − 1, j + 1)
by the inclusion exclusion principle.
Vin de Silva Pomona College 5. Persistence Diagrams
Relationship between the two diagrams
Inverse transformations
• • • • • 4
• • • • 3
• • • 2
• • 1
• 0
0 1 2 3 4 r(i, j)
//oo
• • • • • 4
• • • • 3
• • • 2
• • 1
• 0
0 1 2 3 4 m(i, j)
rank diagram ← persistence diagram
Each summand I (p, q) contributes to r(i , j) whenever p ≤ i ≤ j ≤ q. Thus:
r(i , j) =∑i
p=0
∑nq=j m(p, q)
rank diagram → persistence diagram
The inverse transformation is
m(i , j) = r(i , j)− r(i − 1, j)− r(i , j + 1) + r(i − 1, j + 1)
by the inclusion exclusion principle.
Vin de Silva Pomona College 5. Persistence Diagrams
Relationship between the two diagrams
Inverse transformations
• • • • • 4
• • • • 3
• • • 2
• • 1
• 0
0 1 2 3 4 r(i, j)
//oo
• • • • • 4
• • • • 3
• • • 2
• • 1
• 0
0 1 2 3 4 m(i, j)
rank diagram ← persistence diagram
Each summand I (p, q) contributes to r(i , j) whenever p ≤ i ≤ j ≤ q. Thus:
r(i , j) =∑i
p=0
∑nq=j m(p, q)
rank diagram → persistence diagram
The inverse transformation is
m(i , j) = r(i , j)− r(i − 1, j)− r(i , j + 1) + r(i − 1, j + 1)
by the inclusion exclusion principle.
Vin de Silva Pomona College 5. Persistence Diagrams
Relationship between the two diagrams
Inverse transformations
0 0 0 0 0 4
0 2 2 2 3
1 3 3 2
1 3 1
1 0
0 1 2 3 4 r(i, j)
//oo
◦ ◦ ◦ ◦ ◦ 4
◦ 2 ◦ ◦ 3
1 ◦ ◦ 2
◦ ◦ 1
◦ 0
0 1 2 3 4 m(i, j)
rank diagram ← persistence diagram
Each summand I (p, q) contributes to r(i , j) whenever p ≤ i ≤ j ≤ q. Thus:
r(i , j) =∑i
p=0
∑nq=j m(p, q)
rank diagram → persistence diagram
The inverse transformation is
m(i , j) = r(i , j)− r(i − 1, j)− r(i , j + 1) + r(i − 1, j + 1)
by the inclusion exclusion principle.
Vin de Silva Pomona College 5. Persistence Diagrams
Plotting the persistence diagram
Plotting conventions
For persistence modules such as the Vietoris–Rips homology of a finite metricspace, constructed with respect to a set of critical values
r0 < r1 < · · · < rn
we adopt the following convention. Each summand I (i , j) is drawn as
• a half-open interval [ri , rj+1) in the barcode; or
• a point (ri , rj+1) in the persistence diagram.
We set rn+1 = +∞ by convention.Persistence diagram
smallscale
largescale
Barcode0 0.5 10
0.5
1
small scale
large scale
Persistence diagramintervals [b,d)points (b,d)
Persistence diagram
smallscale
largescale
Barcode0 0.5 10
0.5
1
small scale
large scale
Persistence diagramintervals [b,d)points (b,d)Vin de Silva Pomona College 5. Persistence Diagrams
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Edelsbrunner, Letscher, Zomorodian (2000)
Given a nested sequence of simplicial complexes
S0⊆// S1
⊆// . . .
⊆// Sn
we wish to compute the persistence diagram of
Hd(S0) // Hd(S1) // . . . // Hd(Sn).
We compute it for all d simultaneously (up to some maximum dimension).
Input
A sequence of ‘cells’ σ0, σ1, . . . , σn. For each cell we are given:
• the dimension d = dim(σk);
• the boundary ∂[σk ], as a cycle built from existing (d − 1)-cells.
Output
The persistence diagram for this nested sequence of formal cell-complexes.
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Edelsbrunner, Letscher, Zomorodian (2000)
Given a nested sequence of simplicial complexes
S0⊆// S1
⊆// . . .
⊆// Sn
we wish to compute the persistence diagram of
Hd(S0) // Hd(S1) // . . . // Hd(Sn).
We compute it for all d simultaneously (up to some maximum dimension).
Input
A sequence of ‘cells’ σ0, σ1, . . . , σn. For each cell we are given:
• the dimension d = dim(σk);
• the boundary ∂[σk ], as a cycle built from existing (d − 1)-cells.
Output
The persistence diagram for this nested sequence of formal cell-complexes.
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Data structure
We maintain an upper-triangular square matrix, size equal to the number of cellsprocessed so far. Columns are coloured green, yellow, red.
colour content leading coefficient
green d-cycle αk 1yellow d-cycle βk 1red d-chain γk non-zero
Here d is the dimension of the cell σk whose column it is.
We maintain a bijection between the yellow and red columns.If yellow column k is paired with red column `, then k < ` and
∂γ` = βk .
Readout
Each green column j corresponds to an interval [j ,+∞).Each yellow–red pair (k, `) corresponds to an interval [k, `).
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Matrix
1 2 3 12 23 13
1 1 -1
2 1
3 1
12 ∗ 1
23
13
1
23
Chains
green : α0 = [1], α2 = [3]
yellow : β1 = [2]− [1]
red : γ3 = [1, 2]
Persistence intervals
[0,+∞), [1, 3), [2,+∞)
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Lemma
The green and yellow columns form a basis for the space of all cycles.The yellow columns form a basis for the space of all boundaries.
Cycle reduction
Using the table, any cycle ψ can be uniquely written
ψ = ∂ρ or ψ = ∂ρ+ φ
ρ is a combination of red columns, φ is a cycle whose leading simplex is green.
Iterative method
Initialise ρ = 0.Consider ψ − ∂ρ.
• If it is zero then STOP.
• If its leading simplex is green, then set φ = ψ − ∂ρ and STOP.
• If its leading simplex is yellow, then add to ρ the appropriate multiple ofthe associated red column to eliminate that leading term.
Repeat.
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Lemma
The green and yellow columns form a basis for the space of all cycles.The yellow columns form a basis for the space of all boundaries.
Cycle reduction
Using the table, any cycle ψ can be uniquely written
ψ = ∂ρ or ψ = ∂ρ+ φ
ρ is a combination of red columns, φ is a cycle whose leading simplex is green.
Iterative method
Initialise ρ = 0.Consider ψ − ∂ρ.
• If it is zero then STOP.
• If its leading simplex is green, then set φ = ψ − ∂ρ and STOP.
• If its leading simplex is yellow, then add to ρ the appropriate multiple ofthe associated red column to eliminate that leading term.
Repeat.
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Lemma
The green and yellow columns form a basis for the space of all cycles.The yellow columns form a basis for the space of all boundaries.
Cycle reduction
Using the table, any cycle ψ can be uniquely written
ψ = ∂ρ or ψ = ∂ρ+ φ
ρ is a combination of red columns, φ is a cycle whose leading simplex is green.
Iterative method
Initialise ρ = 0.Consider ψ − ∂ρ.
• If it is zero then STOP.
• If its leading simplex is green, then set φ = ψ − ∂ρ and STOP.
• If its leading simplex is yellow, then add to ρ the appropriate multiple ofthe associated red column to eliminate that leading term.
Repeat.
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Update Step
Consider the next simplex σk .
If ∂[σk ] was already a boundary: The cycle reduction lemma gives
∂[σk ] = ∂ρ.
We get a new green cycleαk = [σk ]− ρ.
If ∂[σk ] was not previously a boundary: The cycle reduction lemma gives
∂[σk ] = ∂ρ+ φ
where the leading term of φ is c[σj ] for some scalar c.Change the j-th column to yellow, and colour the k-th column red. Set
βj = c−1φ, γk = c−1([σk ]− ρ)
(discarding αj). Pair the columns j and k.
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Update Step
Consider the next simplex σk .
If ∂[σk ] was already a boundary: The cycle reduction lemma gives
∂[σk ] = ∂ρ.
We get a new green cycleαk = [σk ]− ρ.
If ∂[σk ] was not previously a boundary: The cycle reduction lemma gives
∂[σk ] = ∂ρ+ φ
where the leading term of φ is c[σj ] for some scalar c.Change the j-th column to yellow, and colour the k-th column red. Set
βj = c−1φ, γk = c−1([σk ]− ρ)
(discarding αj). Pair the columns j and k.
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Update Step
Consider the next simplex σk .
If ∂[σk ] was already a boundary: The cycle reduction lemma gives
∂[σk ] = ∂ρ.
We get a new green cycleαk = [σk ]− ρ.
If ∂[σk ] was not previously a boundary: The cycle reduction lemma gives
∂[σk ] = ∂ρ+ φ
where the leading term of φ is c[σj ] for some scalar c.Change the j-th column to yellow, and colour the k-th column red. Set
βj = c−1φ, γk = c−1([σk ]− ρ)
(discarding αj). Pair the columns j and k.
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Matrix
1 2 3 12 23 13
1 1 -1
2 1
3 1
12 ∗ 1
23
13
1
23
Chains
green : α0 = [1], α2 = [3]
yellow : β1 = [2]− [1]
red : γ3 = [1, 2]
Persistence intervals
[0,+∞), [1, 3), [2,+∞)
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Matrix
1 2 3 12 23 13
1 1 -1
2 1 -1
3 1
12 ∗ 1
23 ∗ 1
13
1
23
Chains
green : α0 = [1]
yellow : β1 = [2]− [1], β2 = [3]− [2]
red : γ3 = [1, 2], γ4 = [2, 3]
Persistence intervals
[0,+∞), [1, 3), [2, 4)
Vin de Silva Pomona College 5. Persistence Diagrams
The ELZ Persistence Algorithm
Matrix
1 2 3 12 23 13
1 1 -1
2 1 -1
3 1
12 ∗ 1 -1
23 ∗ 1 -1
13 1
1
23
Chains
green : α0 = [1], α5 = [1, 3]− [2, 3]− [1, 2]
yellow : β1 = [2]− [1], β2 = [3]− [2]
red : γ3 = [1, 2], γ4 = [2, 3]
Persistence intervals
[0,+∞), [1, 3), [2, 4), [5,+∞)
Vin de Silva Pomona College 5. Persistence Diagrams
Vin de Silva Pomona College 5. Persistence Diagrams