![Page 1: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/1.jpg)
Michael Robinson
Filtrations of covers, Sheaves,
and Integration
![Page 2: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/2.jpg)
Michael Robinson
Acknowledgements● Collaborators:
– Brett Jefferson, Cliff Joslyn, Brenda Praggastis, Emilie Purvine (PNNL)
– Chris Capraro, Grant Clarke, Griffin Kearney, Janelle Henrich, Kevin Palmowski (SRC)
– J Smart, Dave Bridgeland (Georgetown)● Students:
– Philip Coyle– Samara Fantie
● Recent funding: DARPA, ONR, AFRL
– Robby Green– Fangei Lan
– Michael Rawson– Metin Toksoz-Exley– Jackson Williams
![Page 3: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/3.jpg)
Michael Robinson
Key ideas● Motivate filtrations of partial covers as
generalizing consistency filtrations of sheaf assignments
● Explore filtrations of partial covers as interesting mathematical objects in their own right
● Instill hope that filtrations of partial covers may be (incompletely) characterized
Big caveat: Only finite spaces are under consideration!
![Page 4: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/4.jpg)
Michael Robinson
Consistency filtrations
![Page 5: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/5.jpg)
Michael Robinson
Context● Assemble stochastic models of data locally into a
global topological picture– Persistent homology is sensitive to outliers– Statistical tools are less sensitive to outliers, but cannot
handle (much) global topological structure– Sheaves can be built to mediate between these two
extremes... this is what I have tried to do for the past decade or so
● The output is the consistency filtration of a sheaf assignment
![Page 6: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/6.jpg)
Michael Robinson
Topologizing a partial order
Open sets are unionsof up-sets
![Page 7: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/7.jpg)
Michael Robinson
Topologizing a partial order
Intersectionsof up-sets are alsoup-sets
![Page 8: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/8.jpg)
Michael Robinson
( )( )
A sheaf on a poset is...
A set assigned to each element, calleda stalk, and …
ℝ
ℝ2 ℝ2
ℝ3
0 1 11 0 1
1 0 10 1 1
(1 0)(0 1)
ℝ ℝ2
ℝ2
(1 -1) ( )0 11 0
( )-3 3-4 4
This is a sheaf of vector spaces on a partial order
ℝ3
( )0 1 11 0 1
ℝ
(-2 1)
Stalks can be measure spaces!We can handle stochastic data
231
2 -23 -31 -1
![Page 9: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/9.jpg)
Michael Robinson
( )( )
A sheaf on a poset is...
… restriction functions between stalks, following the order relation…
ℝ
ℝ2 ℝ2
ℝ3
0 1 11 0 1
1 0 10 1 1
(1 0)(0 1)
ℝ ℝ2
ℝ2
(1 -1)
231
( )0 11 0
( )-3 3-4 4
2 -23 -31 -1
This is a sheaf of vector spaces on a partial order
ℝ3
( )0 1 11 0 1
ℝ
(-2 1)
(“Restriction” because it goes frombigger up-sets to smaller ones)
![Page 10: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/10.jpg)
Michael Robinson
( )( )
A sheaf on a poset is...ℝ
ℝ2 ℝ2
ℝ3
0 1 11 0 1
1 0 10 1 1
(1 0)(0 1)
ℝ ℝ2
ℝ2
(1 -1) ( )0 11 0
( )-3 3-4 4
This is a sheaf of vector spaces on a partial order
ℝ3
( )0 1 11 0 1
ℝ
(-2 1)
(1 -1) =
( )0 1 11 0 1(1 0) = (0 1) ( )1 0 1
0 1 1
( )0 11 0( )-3 3
-4 4( )1 0 10 1 1
2 -23 -31 -1
=
… so that the diagramcommutes!
231
231
2 -23 -31 -1
2 -23 -31 -1
![Page 11: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/11.jpg)
Michael Robinson
( )( )
An assignment is...
… the selection of avalue on some open sets
0 1 11 0 1
1 0 10 1 1
(1 0)(0 1)
(1 -1)
231
( )0 11 0
( )-3 3-4 4
2 -23 -31 -1 ( )0 1 1
1 0 1
(-2 1)
(-1)
( )23
( )-4-3
( )-3-4
( )32
(-4)
(-4)
230
-2-3-1
The term serration is more common, but perhaps more opaque.
![Page 12: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/12.jpg)
Michael Robinson
( )( )
A global section is...
… an assignmentthat is consistent with the restrictions
0 1 11 0 1
1 0 10 1 1
(1 0)(0 1)
(1 -1)
231
( )0 11 0
( )-3 3-4 4
2 -23 -31 -1 ( )0 1 1
1 0 1
(-2 1)
(-1)
( )23
( )-4-3
( )-3-4
( )32
(-4)
(-4)
230
-2-3-1
![Page 13: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/13.jpg)
Michael Robinson
( )( )
Some assignments aren’t consistent
… but they mightbe partially consistent
0 1 11 0 1
1 0 10 1 1
(1 0)(0 1)
(1 -1)
231
( )0 11 0
( )-3 3-4 4
2 -23 -31 -1 ( )0 1 1
1 0 1
(-2 1)
(+1)
( )23
( )-4-3
( )-3-4
( )32
(-4)
(-4)
231
-2-3-1
![Page 14: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/14.jpg)
Michael Robinson
( )( )
Consistency radius is...… the maximum (or some other norm)distance between the value in a stalk and the values propagated along the restrictions
0 1 11 0 1
1 0 10 1 1
(1 0)(0 1)
(1 -1)
231
( )0 11 0
( )-3 3-4 4
2 -23 -31 -1 ( )0 1 1
1 0 1
(-2 1)
(+1)
( )23
( )-4-3
( )-3-4
( )32
(-4)
(-4)
231
-2-3-1
231( )0 1 1
1 0 1 ( )32- = 2
( )23(1 -1) - 1 = 2
(+1) - 231
-2-3 = 2 14-1 MAX ≥ 2 14
Note: lots more restrictions to check!
![Page 15: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/15.jpg)
Michael Robinson
Amateur radio foxhunting
Typical sensors:● Bearing to Fox● Fox signal strength● GPS location
![Page 16: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/16.jpg)
Michael Robinson
Bearing sensors
0.5
1
0
30
60
90
120
150
180
210
240
270
300
330
Antenna pattern
![Page 17: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/17.jpg)
Michael Robinson
Bearing sensors… reality…
0.5
1
0
30
60
90
120
150
180
210
240
270
300
330
Antenna pattern
![Page 18: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/18.jpg)
Michael Robinson
Bearing observations
Bearing as a function of sensor position
0.5
1
0
30
60
90
120
150
180
210
240
270
300
330
Antenna pattern
Recenter
![Page 19: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/19.jpg)
Michael Robinson
Bearing sheaf
Typical Mbearing function:
0.5
1
0
30
60
90
120
150
180
210
240
270
300
330
Antenna pattern
![Page 20: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/20.jpg)
Michael Robinson
Bearing sheaf (two sensors)Fox position Sensor 2 position, BearingSensor 1 position, Bearing
Fox position, Sensor 1 position Fox position, Sensor 2 positionℝ2×ℝ2
ℝ2×S1 ℝ2×S1
ℝ2×ℝ2
ℝ2
pr1pr1(pr2,Mbearing)
(pr2,Mbearing)
Global sections of this sheaf correspond to two bearings whose sight lines intersect at the fox transmitter
![Page 21: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/21.jpg)
Michael Robinson
Consistency of proposed fox locationsConsistency radius minimization …
… converges to a likely fox location … does not converge!
![Page 22: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/22.jpg)
Michael Robinson
Local consistency radius
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
Consistency radius of this open set = 0
Lemma: Consistency radius on an open set U is computed by only considering open sets V1 ⊆ V2 ⊆ U
{A,C} {B,C}
{C}
{A,B,C}
![Page 23: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/23.jpg)
Michael Robinson
Local consistency radius
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
c(U) = ½
Lemma: Consistency radius does not decrease as its support grows:if U ⊆ V then c(U) ≤ c(V).
Consistency radius of this open set = 0
![Page 24: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/24.jpg)
Michael Robinson
Local consistency radius
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
c(U) = ½ c(V) = ½
c(U ∩ V) = 0
Lemma: Consistency radius does not decrease as its support grows:if U ⊆ V then c(U) ≤ c(V).
![Page 25: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/25.jpg)
Michael Robinson
Consistency radius is not a measure
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
c(U) = ½ c(V) = ½
c(U ∩ V) = 0
c(U ∪ V) = ⅔ ≠ c(U) + c(V) – c(U ∩ V)
(Consistency radius yields an inner measure after some work)
![Page 26: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/26.jpg)
Michael Robinson
The consistency filtration
1
⅓
(1)
(1)
Consistencythreshold 0 ½ ⅔
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
Consistency radius = ⅔
● … assigns the set of open sets (open cover) with consistency less than a given threshold
● Lemma: consistency filtration is itself a sheaf of collections of open sets on (ℝ,≤). Restrictions in this sheaf are cover coarsenings.
refine refine
![Page 27: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/27.jpg)
Michael Robinson
Filtrations of partial covers
![Page 28: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/28.jpg)
Michael Robinson
Covers of topological spaces● Classic tool: Čech cohomology
– Coarse– Usually blind to the cover; only sees the underlying space
● Cover measures (Purvine, Pogel, Joslyn, 2017)– How fine is a cover?– How overlappy is a cover?
![Page 29: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/29.jpg)
Michael Robinson
Cover measures● Theorem: (Birkhoff) The set of covers ordered by
refinement has an explicit rank function– The rank of a given cover is the number of sets in its
downset as an antichain of the Boolean lattice– This counts the number of sets of consistent faces there are
● Conclusion: An assignment whose maximal cover has a higher rank is more self-consistent
![Page 30: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/30.jpg)
Michael Robinson
Cover measures● Consider the following two covers of {1,2,3,4}
1 2 3 4A
1 2 3 4B
1 2 3 4 1 2 3 4
1 2 3 4
Refine Refine
RefineTotal = 6 sets
Total = 11 sets
Since 6 < 11, cover B is coarser
![Page 31: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/31.jpg)
Michael Robinson
The lattice of covers● Theorem: The lattice of
covers is graded using this rank function
1 2 3 4
1 2 3 4
1 2 3 4
1 2 3 4
Coarser
Finer
Lattice graphic by E. Purvine
![Page 32: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/32.jpg)
Michael Robinson
Defining CTop : partial covers● Start with a fixed topological space● Objects: Collections of open sets● No requirement of coverage 1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
![Page 33: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/33.jpg)
Michael Robinson
Defining CTop : partial covers● Morphisms are refinements of
covers:
If 𝒰 and 𝒱 are partial covers, 𝒱 refines 𝒰 if for all V in 𝒱 there is a U in 𝒰, with V ⊆ U.
● Convention: 𝒰 → 𝒱
1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
![Page 34: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/34.jpg)
Michael Robinson
Irredundancy● Irredundant cover has no cover
elements contained in others● Minimal representatives of
CTop isomorphism classes– According to inclusion, not
refinement● Lemma: Every finite partial
cover is CTop-isomorphic to a unique irredundant one
1 2 3 4 5 6Refine
1 2 3 4 5 6
![Page 35: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/35.jpg)
Michael Robinson
Defining SCTop : Filtrations in CTop● Objects are chains of morphisms in CTop with a
monotonic height function– Height increases as cover coarsens– Could be the cover lattice rank,
but need not be 1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
1.0
0.5
0.0
height
![Page 36: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/36.jpg)
Michael Robinson
Defining SCTop : Filtrations in CTop● Morphisms are commutative ladders of refinements
with a monotonic mapping ϕ : ℝ→ℝ of height functions
1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
1.0
0.5
0.0
height
1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
1.2
0.7
0.3
0.1height 1 2 3 4 5 6
Refine
![Page 37: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/37.jpg)
Michael Robinson
Defining SCTop : Filtrations in CTop● Morphisms are commutative ladders of refinements
with a monotonic mapping ϕ : ℝ→ℝ of height functions
1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
1.0
0.5
0.0
height
1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
1.2
0.7
0.3
0.1height 1 2 3 4 5 6
Refine
![Page 38: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/38.jpg)
Michael Robinson
Defining SCTop : Filtrations in CTop● Morphisms are commutative ladders of refinements
with a monotonic mapping ϕ : ℝ→ℝ of height functions
1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
1.0
0.5
0.0
height
1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
1.2
0.7
0.3
0.1height 1 2 3 4 5 6
Refine
Refine
Refine
Refine
![Page 39: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/39.jpg)
Michael Robinson
Interleavings in SCTop● Pair of morphisms between two objects
1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
1.0
0.5
0.0
height
1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
1.2
0.7
0.3
0.1height 1 2 3 4 5 6
Refine
Refine
Refine
Refine
![Page 40: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/40.jpg)
Michael Robinson
Interleavings in SCTop● Measure the maximum displacement of the heights,
minimize over all interleavings = interleaving distance
1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
1.0
0.5
0.0
height
1 2 3 4 5 6
1 2 3 4 5 6
1 2 3 4 5 6
Refine
Refine
1.2
0.7
0.3
0.1height 1 2 3 4 5 6
Refine
Refine
Refine
Refine
Dist = 0.2
Dist = 0.3
Dist = 0.2
Dist = 0.2
Dist = 0.2
Dist = 0.3
Dist = 0.2
Dist = 0.1
![Page 41: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/41.jpg)
Michael Robinson
Consistency filtration stability
1
⅓
(1)
(1)
0 ½ ⅔
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
Consistency radius = ⅔
● Theorem: Consistency filtration is continuous under an appropriate interleaving distance
● Thus the persistent Čech cohomology of the consistency filtration is robust to perturbations
Consistencythreshold
refine refine
![Page 42: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/42.jpg)
Michael Robinson
A small perturbation … ● Perturbations allowed in both assignment and
sheaf (subject to it staying a sheaf!)
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
0 0.9
0.6
0.3(2)
(1)
(0.8)
(0.8)
(0.4)Max difference = 0.2
![Page 43: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/43.jpg)
Michael Robinson
A small perturbation … ● Compute consistency filtrations...
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
0 ½ ⅔Consistencythreshold
{C}
{A,C}
{B,C}
{A,B,C}
0 0.9
0.6
0.3(2)
(1)
(0.8)
(0.8)
(0.4)
0 0.6 0.66Consistencythreshold
{C}
{A,C}
{B,C}
{A,B,C}
0.3
Max difference = 0.2
![Page 44: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/44.jpg)
Michael Robinson
… bounds interleaving distance
0 ½ ⅔
Consistencythreshold
{C}
{A,C}
{B,C}
{A,B,C}
0 0.6 0.66
Consistencythreshold
{C}
{A,C}
{B,C}
{A,B,C}
0.3
{C}
{A,C}
{B,C}
{A,B,C}
Shift = 0.5-0.3 = 0.2
Refine Note that {A,C} ⊆ {A,B,C} etc.
![Page 45: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/45.jpg)
Michael Robinson
… bounds interleaving distance
0 ½ ⅔
Consistencythreshold
{C}
{A,C}
{B,C}
{A,B,C}
0 0.6 0.66
Consistencythreshold
{C}
{A,C}
{B,C}
{A,B,C}
0.3
{C}
{A,C}
{B,C}
{A,B,C}
Shift = 0.6-0.5 = 0.1
Max shift = 0.2, This is bounded above by constant times the perturbation (0.2 in this case)Refine
![Page 46: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/46.jpg)
Michael Robinson
Summarizing a filtration
![Page 47: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/47.jpg)
Michael Robinson
Defining Con : consistency functions● Objects: order preserving functions Open(X) → ℝ+
● Example: local consistency radius
0 1
½
⅓(2)
(1)
(1)
(1)
(½)
{A,C} {B,C}
{C}
{A,B,C}
½ ½
0
⅔
Open sets Assignment Local consistency radiusObject in Con
![Page 48: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/48.jpg)
Michael Robinson
Defining Con : consistency functionsIdeally, we want…● Consistency radius is a functor ShvFPA → Con● A functor Con → SCTop acting by thresholding
{A,C} {B,C}
{C}
{A,B,C}
½ ½
0
⅔
Open sets Local consistency radiusObject in Con
A C B
A C B
A C B
Refine
Refine
Consistency filtrationObject in SCTop
Hei
ght =
con
sist
ency
![Page 49: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/49.jpg)
Michael Robinson
Defining Con : consistency functionsIdeally, we want…● Consistency radius is a functor ShvFPA → Con● A functor Con → SCTop acting by thresholding
To get this, the morphisms of Con are a little strange
A morphism K: m → n of Con is a nonnegative real K so that m(U) ≤ K n(U) for all open U.
Composition works by multiplication!
![Page 50: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/50.jpg)
Michael Robinson
Defining Con : consistency functionsA morphism K: m → n of Con is a nonnegative real K so that m(U) ≤ K n(U) for all open U.
½ ½
0
⅔
Object in Con
½
0
⅔
Object in Con
1
1
½
2
½
These objects are not Con-isomorphic!
![Page 51: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/51.jpg)
Michael Robinson
Con and SCTopTheorem: Con is equivalent to a subcategory of SCTop by way of two functors:● A faithful functor Con → SCTop● A non-faithful functor SCTop → Con
such that Con → SCTop → Con is the identity functor.
Interpretation: May be able to summarize filtrations of partial covers using consistency functions, but this is lossy!
![Page 52: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/52.jpg)
Michael Robinson
Con → SCTop● Motivation: generalization of consistency filtration● Idea: thresholding!
0
⅔
Object in Con
1
A C B
A C B
Refine
Refine
Object in SCTop
A C B
A C B
Refine½½
0
⅔
1
Showing an irredundant representative for this object{A,C} {B,C}
{C}
{A,B,C}
Open sets
![Page 53: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/53.jpg)
Michael Robinson
Con → SCTop● Morphisms in Con transform to linear rescalings
of the heights in SCTop … monotonicity does the rest
½
0
⅔
Morphism in Con
1
A C B
A C B
Refine
Refine
Morphism in SCTop
A C B
A C B
Refine
0
½
⅔
1
½ ½
0
⅔
2
A C B
A C B
A C B
Refine
Refine
½
0
⅔
1
![Page 54: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/54.jpg)
Michael Robinson
Con → SCTop● Morphisms in Con transform to linear rescalings
of the heights in SCTop … monotonicity does the rest
½
0
⅔
Morphism in Con
1
A C B
A C B
Refine
Refine
Morphism in SCTop
A C B
A C B
Refine
0
½
⅔
1
½ ½
0
⅔
2
A C B
A C B
A C B
Refine
Refine
½
0
⅔
1
![Page 55: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/55.jpg)
Michael Robinson
Con → SCTop● Morphisms in Con transform to linear rescalings
of the heights in SCTop … monotonicity does the rest: Faithful!
½
0
⅔
Morphism in Con
1
A C B
A C B
Refine
Refine
Morphism in SCTop
A C B
A C B
Refine
0
½
⅔
1
½ ½
0
⅔
2
A C B
A C B
A C B
Refine
Refine
½
0
⅔
1
![Page 56: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/56.jpg)
Michael Robinson
SCTop → Con● At first, this seems easy. Just look up the threshold
for each open setA C B
A C B
Refine
Refine
Object in SCTop
A C B
A C B
Refine
½
0
⅔
1
0
⅔
Object in Con
1
?{A,C} {B,C}
{C}
{A,B,C}
Open sets
![Page 57: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/57.jpg)
Michael Robinson
SCTop → Con● At first, this seems easy. Just look up the threshold
for each open setA C B
A C B
Refine
Refine
Object in SCTop
A C B
A C B
Refine
½
0
⅔
1
0
⅔
Object in Con
1
½{A,C} {B,C}
{C}
{A,B,C}
Open sets
![Page 58: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/58.jpg)
Michael Robinson
SCTop → Con● But what if the cover is not irredundant?● This does not matter!
0
⅔
Object in Con
1
A C B
Refine
Refine
Object in SCTop
Refine½½
0
⅔
1
A C B
A C B
A C B
Fix: take the smallest threshold where the open set is contained in a cover element
![Page 59: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/59.jpg)
Michael Robinson
SCTop → Con
½
A C B
A C B
Refine
Refine
Morphism in SCTop
A C B
A C B
Refine
0
⅔
1
A C B
A C B
A C B
Refine
Refine
½
0
⅔
1
0
⅔
Morphism in Con
1
½ ½
0
⅔
● Recall: SCTop morphisms are given by height rescaling functions ϕ, which may not be linear
![Page 60: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/60.jpg)
Michael Robinson
SCTop → Con● Morphism in Con is given by K = max ● Not faithful!
½
A C B
A C B
Refine
Refine
Morphism in SCTop
A C B
A C B
Refine
0
⅔
1
A C B
A C B
A C B
Refine
Refine
½
0
⅔
1
0
⅔
Morphism in Con
1
½ ½
0
⅔
tϕ(t)
2
![Page 61: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/61.jpg)
Michael Robinson
Hope for a characterization
![Page 62: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/62.jpg)
Michael Robinson
Pivoting to inner measures● SCTop and Con are fairly elaborate
– Objects are “two-level”: open subsets, labeled with real values
● It’s potentially useful to study their interaction with functions on the underlying space– The way to do this is via integration– But then we need to transform Con objects into something
like a measure… they have no hope of being additive!● One way to do that is through an endofunctor that
creates inner measures, Inner : Con→Con
![Page 63: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/63.jpg)
Michael Robinson
Inner measures from ConTheorem: Suppose that m is an object in Con, a consistency function. A Borel inner measure is generated by
Im(V) = m(U).
An inner measure satisfies● I(∅) = 0● I(U) ≥ 0● I(U ∪ V) ≥ I(U) + I(V) – I(U ∩ V)
U ⊆ VΣ
Note: this is an endofunctor since if m(U) ≤ K n(U), then Im(U) ≤ K In(U) by linearity
![Page 64: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/64.jpg)
Michael Robinson
Integration w.r.t. inner measures● This works like Baryshnikov-Ghrist real-valued Euler
integration (strong connection to sheaves!)● For an f : X → ℝ and inner measure I, define
f dI = lim [I(f-1([z/n,∞))) – I(f-1((-∞,-z/n]))]
● Theorem: (Monotonicity) 0 ≤ f ≤ g implies f dI ≤ g dI● Theorem: (Partial linearity) (r f) dI = r f dI
and if 0 ≤ f ≤ g, (f + g) dI ≥ f dI + g dI
Σ 1n
z = 1
∞
n→∞∫
Sublevel setsSuperlevel setsResolution
∫ ∫∫∫
∫ ∫ ∫
![Page 65: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/65.jpg)
Michael Robinson
Missing structureCannot remove the inequality, though. There are functions f, g for which (f + g) dI ≠ f dI + g dI
Additionally,● (f g) dI is not an inner product
● | f |p dI is not a norm, and
● | f – g |p dI 1/p is not a pseudometric.
Is there a topology induced by an inner measure?
∫ ∫ ∫
∫
∫
∫
![Page 66: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/66.jpg)
Michael Robinson
Open questions...● What’s the interaction between topological
invariants (Euler characteristic/integral), geometric invariants inner measures, and filtrations?– Expect a sheaf-theoretic answer!
● How robust are inner measures to distortions or stochastic variability?– Use interleaving distance on SCTop!– How does that push forward to inner measures?
● What does the integral of a function on a base space of a sheaf assignment mean?
![Page 67: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/67.jpg)
Michael Robinson
More open questions...● Can we relate structure of local consistency of a
sheaf assignment to the structure of functions on the base?
● Do all inner measures arise from consistency functions?– They definitely don’t have to come from sheaf
assignments!● What is the most natural topology on the space of
inner measures?● Is there a better notion of measures for consistency
functions? Hopefully an actual measure?
![Page 68: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/68.jpg)
Michael Robinson
To learn more...Michael Robinson
http://drmichaelrobinson.net
Preprint: https://arxiv.org/abs/1805.08927
Software: https://github.com/kb1dds/pysheaf
![Page 69: Filtrations of covers, Sheaves, and Integration...Michael Robinson Context Assemble stochastic models of data locally into a global topological picture – Persistent homology is sensitive](https://reader033.vdocuments.us/reader033/viewer/2022042923/5f700734e5a7356a1b75cf53/html5/thumbnails/69.jpg)
Michael Robinson
Excursion! Friday evening ~ 6pm● RetroComputing Society of Rhode Island