Products in the category of approach spaces as models for complexity - - PowerPoint PPT Presentation

products in the category of approach spaces as models for
SMART_READER_LITE
LIVE PREVIEW

Products in the category of approach spaces as models for complexity - - PowerPoint PPT Presentation

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit Products in the category of approach spaces as models for complexity Eva Colebunders Vrije Universiteit


slide-1
SLIDE 1

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Products in the category of approach spaces as models for complexity

Eva Colebunders

Vrije Universiteit Brussel

Workshop on Category Theory, Coimbra, July 2012 Conference in honor of George Janelidze, on the occasion of his 60-th birthday

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-2
SLIDE 2

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Complexity of algorithms

Complexity of certain types of algorithms, is described as a solution of some recurrence equation. For Mergesort the running time f : N →]0, ∞] is a solution of the equation:

  • f (n) = c for n = 1

f (n) = a.f [ n

b] + h(n) whenever n = 1

for given a, b, c and h : N →]0, ∞].

  • M. P. Schellekens, The Smyth completion: A common foundation

for denotational semantics and complexity analysis, Elect. Notes

  • Theoret. Comp. Sci., (1995).

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-3
SLIDE 3

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Complexity of algorithms 2

Calculations of running time of other examples like Quicksort fit into the following recurrence equation:

  • f (n) = cn for 1 ≤ n ≤ k

f (n) = Σi=k

i=1ai.f (n − i) + h(n) whenever n > k

for given k, ai and h : N →]0, ∞]

  • S. Romaguera and O. Valero, A common Mathematical Framework

for Asymptotic Complexity Analysis and Denotational Semantics for Recursive Programs Based on Complexity spaces, International Journal of Computer Mathematics, 2012.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-4
SLIDE 4

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Associated fixed point problem

Reformulating the problem as a fixed point result: X =]0, ∞]N and Φ : X → X : g → Φg.

  • Φg(n) = cn for 1 ≤ n ≤ k

Φg(n) = Σi=k

i=1ai.g(n − i) + h(n) whenever n > k

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-5
SLIDE 5

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Other references

  • S. Romaguera, M.P. Schellekens, P. Tirado, O. Valero,

Contraction selfmaps on complexity spaces and ExpoDC algorithms, Amer. Inst. Physics Proceedings, (2007).

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-6
SLIDE 6

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Other references

  • S. Romaguera, M.P. Schellekens, P. Tirado, O. Valero,

Contraction selfmaps on complexity spaces and ExpoDC algorithms, Amer. Inst. Physics Proceedings, (2007).

  • L. M. Garc´

ıa-Raffi, S. Romaguera, M. P. Schellekens, Applications of the complexity space to the general probabilistic divide and conquer algorithms,

  • J. Math. Anal. Appl. (2008).

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-7
SLIDE 7

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Solutions of the fixed point problems

Method The complexity distance. dC(f , g) = Σn∈N 1 2n .[( 1 g(n) − 1 f (n)) ∨ 0]

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-8
SLIDE 8

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Solutions of the fixed point problems

Method The complexity distance. dC(f , g) = Σn∈N 1 2n .[( 1 g(n) − 1 f (n)) ∨ 0] On C = {g ∈ ]0, ∞]N | Σn∈N 1 2n . 1 g(n) < ∞}.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-9
SLIDE 9

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Solutions of the fixed point problems

Method The complexity distance. dC(f , g) = Σn∈N 1 2n .[( 1 g(n) − 1 f (n)) ∨ 0] On C = {g ∈ ]0, ∞]N | Σn∈N 1 2n . 1 g(n) < ∞}. (C, dC) is bicomplete

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-10
SLIDE 10

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Solutions of the fixed point problems

Method The complexity distance. dC(f , g) = Σn∈N 1 2n .[( 1 g(n) − 1 f (n)) ∨ 0] On C = {g ∈ ]0, ∞]N | Σn∈N 1 2n . 1 g(n) < ∞}. (C, dC) is bicomplete Restrict to Φ : C → C, dC-Lipschitz with factor strictly smaller than 1

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-11
SLIDE 11

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Solutions of the fixed point problems

Method The complexity distance. dC(f , g) = Σn∈N 1 2n .[( 1 g(n) − 1 f (n)) ∨ 0] On C = {g ∈ ]0, ∞]N | Σn∈N 1 2n . 1 g(n) < ∞}. (C, dC) is bicomplete Restrict to Φ : C → C, dC-Lipschitz with factor strictly smaller than 1 Apply the Banach fixed point theorem for quasi metric spaces to obtain a unique fixed point for Φ : C → C.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-12
SLIDE 12

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Our purpose

Results: Changing the categorical context → develop a method applicable to a larger class of recursive algorithms, containing all the previous examples.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-13
SLIDE 13

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Our purpose

Results: Changing the categorical context → develop a method applicable to a larger class of recursive algorithms, containing all the previous examples. Construct App of approach spaces and contractions as morphisms.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-14
SLIDE 14

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Our purpose

Results: Changing the categorical context → develop a method applicable to a larger class of recursive algorithms, containing all the previous examples. Construct App of approach spaces and contractions as morphisms. Categorical product in App → complexity approach space ]0, ∞]N, compatibility with the product in Top and with the pointwise order.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-15
SLIDE 15

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Approach spaces 1

From Top to App: Convergence in an approach space X is described by means of a limit operator.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-16
SLIDE 16

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Approach spaces 1

From Top to App: Convergence in an approach space X is described by means of a limit operator. Objects (X, λ) with λ : FX → [0, ∞]X : F → λF satisfying suitable axioms. A map f : (X, λX) → (Y , λY ) is a contraction if λY (stackf (F)) ◦ f ≤ λXF for every F ∈ F(X).

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-17
SLIDE 17

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Approach spaces 1

From Top to App: Convergence in an approach space X is described by means of a limit operator. Objects (X, λ) with λ : FX → [0, ∞]X : F → λF satisfying suitable axioms. A map f : (X, λX) → (Y , λY ) is a contraction if λY (stackf (F)) ◦ f ≤ λXF for every F ∈ F(X). Top → App is a concretely coreflective full embedding. Given X = (X, λ) we denote its topological coreflection as (X, TX), with TX defined by F → x ⇔ λF(x) = 0.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-18
SLIDE 18

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Approach spaces 2

From qMet to App: Instead of working with one quasi metric we consider a collection of quasi metrics, called a gauge of quasi metrics.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-19
SLIDE 19

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Approach spaces 2

From qMet to App: Instead of working with one quasi metric we consider a collection of quasi metrics, called a gauge of quasi metrics. Objects are (X, G) with G an ideal of quasi metrics d : X × X → [0, ∞] satisfying a certain saturation condition. A map f : (X, GX) → (Y , GY ) is a contraction if ∀q ∈ GY : q ◦ (f × f ) ∈ GX.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-20
SLIDE 20

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Approach spaces 2

From qMet to App: Instead of working with one quasi metric we consider a collection of quasi metrics, called a gauge of quasi metrics. Objects are (X, G) with G an ideal of quasi metrics d : X × X → [0, ∞] satisfying a certain saturation condition. A map f : (X, GX) → (Y , GY ) is a contraction if ∀q ∈ GY : q ◦ (f × f ) ∈ GX. qMet → App is a concretely coreflective full embedding. Given X = (X, G) we denote its quasi metric coreflection as (X, dX), with dX defined as dX = sup G.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-21
SLIDE 21

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

The category of Approach spaces 1

The two constructs are concretely isomorphic. The transition from gauges to limit operators: For an approach space with gauge G, for a filter F and x ∈ X the limit operator is: λF(x) = sup

q∈G

inf

F∈F sup y∈F

q(x, y).

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-22
SLIDE 22

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

The category of Approach spaces 1

The two constructs are concretely isomorphic. The transition from gauges to limit operators: For an approach space with gauge G, for a filter F and x ∈ X the limit operator is: λF(x) = sup

q∈G

inf

F∈F sup y∈F

q(x, y). In particular: for a quasi metric space (X, q) with gauge {d | d ≤ q} the limit operator of a sequence (xn)n is: λ(xn)n(x) = limsupn→∞ q(x, xn)

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-23
SLIDE 23

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

The category of Approach spaces 2

App is a topological construct: Structured source (fi : X → (Xi, λi))i∈I, initial limit operator on F ∈ FX: λF = supi∈Iλi(stackfi(F)) ◦ fi

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-24
SLIDE 24

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Approach complexity spaces 1

Structure X = Z Y : Z = (]0, ∞], ≤) or Z = ([0, ∞], ≤), dcpo for the usual order. p the quasi metric defined by p(x, y) = (1 y − 1 x ) ∨ 0. induces the Scott topology σ(Z, ≤).

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-25
SLIDE 25

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Approach complexity spaces 1

Structure X = Z Y : Z = (]0, ∞], ≤) or Z = ([0, ∞], ≤), dcpo for the usual order. p the quasi metric defined by p(x, y) = (1 y − 1 x ) ∨ 0. induces the Scott topology σ(Z, ≤). For Y arbitrary, X = Z Y is a dcpo for the pointwise order ≤ . Endow X with the product in App, i.e. the initial lift of the source (pry : X → (Z, p))y∈Y . The space X = (X, λ) is called the complexity approach space.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-26
SLIDE 26

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

The limit operator for a sequence (gk)k and f ∈ X is λ(gk)k(f ) = supy∈Y λp(gk(y))k(f (y))

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-27
SLIDE 27

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

The limit operator for a sequence (gk)k and f ∈ X is λ(gk)k(f ) = supy∈Y λp(gk(y))k(f (y)) = supy∈Y limsupk→∞p(f (y), gk(y))

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-28
SLIDE 28

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

The limit operator for a sequence (gk)k and f ∈ X is λ(gk)k(f ) = supy∈Y λp(gk(y))k(f (y)) = supy∈Y limsupk→∞p(f (y), gk(y)) = supy∈Y limsupk→∞( 1 gk(y) − 1 f (y)) ∨ 0 λ(gk)k(f ) ≤ α ⇔ ∀y ∈ Y , ∀η > 0, ∃jy, ∀k ≥ jy ( 1 gk(y) − 1 f (y)) ∨ 0 < α + η

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-29
SLIDE 29

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Fixed points 1

X = Z Y , suppose Y is endowed with a binary irreflexive relation ≺ and for y ∈ Y let Yy = {u ∈ Y | u ≺ y} the initial segment of y. We assume (Y , ≺) is well founded.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-30
SLIDE 30

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Fixed points 1

X = Z Y , suppose Y is endowed with a binary irreflexive relation ≺ and for y ∈ Y let Yy = {u ∈ Y | u ≺ y} the initial segment of y. We assume (Y , ≺) is well founded. Assume: all initial segments Yy = {u ∈ Y | u ≺ y} with y ∈ Y , are finite.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-31
SLIDE 31

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Fixed points 1

X = Z Y , suppose Y is endowed with a binary irreflexive relation ≺ and for y ∈ Y let Yy = {u ∈ Y | u ≺ y} the initial segment of y. We assume (Y , ≺) is well founded. Assume: all initial segments Yy = {u ∈ Y | u ≺ y} with y ∈ Y , are finite. Examples for Y : (N, <) where < is the strict relation associated to the usual well order finite powers Nn, with the “strict pointwise relation” (ni)i ≺ (mi)i if and only if ni < mi for i = 1, · · · , n.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-32
SLIDE 32

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Fixed points 2

X = Z Y is the complexity approach space, and Φ : X → X is of the following type there exists h ∈ X taking only finite values for every y ∈ Y , not minimal, there exists Ψy : Z Yy → Z, such that Φ satisfies Φg(y) =

  • h(y) + Ψy((g(u))u∈Yy )

y not minimal h(y) y minimal for g ∈ X.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-33
SLIDE 33

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Sufficient conditions

For y ∈ Y , y not minimal Z Yy has pointwise order and addition inherited from Z. For a ∈ Z Yy with au = s for every u ∈ Yy we write (au)u = s.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-34
SLIDE 34

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Sufficient conditions

For y ∈ Y , y not minimal Z Yy has pointwise order and addition inherited from Z. For a ∈ Z Yy with au = s for every u ∈ Yy we write (au)u = s. Ψy : Z Yy → Z satisfies the following conditions:

1 Monotone: a ≤ b in Z Yy ⇒ Ψy(a) ≤ Ψy(b) 2 Subadditive: Ψy(a + b) ≤ Ψy(a) + Ψy(b) 3 Limit: ∀ε > 0 ∃δ > 0 : Ψy(s) ≤ ε whenever s ≤ δ 4 Finiteness: If a ∈ Z Yy has only finite coordinates then Ψy(a)

is finite.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-35
SLIDE 35

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Results

Φg(y) =

  • h(y) + Ψy((g(u))u∈Yy )

y not minimal h(y) y minimal Φ has at most one fixed point

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-36
SLIDE 36

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Results

Φg(y) =

  • h(y) + Ψy((g(u))u∈Yy )

y not minimal h(y) y minimal Φ has at most one fixed point Φ : X → X is monotone

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-37
SLIDE 37

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Results

Φg(y) =

  • h(y) + Ψy((g(u))u∈Yy )

y not minimal h(y) y minimal Φ has at most one fixed point Φ : X → X is monotone the function h satisfies h ≤ Φ(h)

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-38
SLIDE 38

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Results

Φg(y) =

  • h(y) + Ψy((g(u))u∈Yy )

y not minimal h(y) y minimal Φ has at most one fixed point Φ : X → X is monotone the function h satisfies h ≤ Φ(h) the sequence (Φkh)k is monotone increasing

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-39
SLIDE 39

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Results

Φg(y) =

  • h(y) + Ψy((g(u))u∈Yy )

y not minimal h(y) y minimal Φ has at most one fixed point Φ : X → X is monotone the function h satisfies h ≤ Φ(h) the sequence (Φkh)k is monotone increasing f =

k Φk(h) exists in X and satisfies f ≤ Φ(f )

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-40
SLIDE 40

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Results

Φg(y) =

  • h(y) + Ψy((g(u))u∈Yy )

y not minimal h(y) y minimal Φ has at most one fixed point Φ : X → X is monotone the function h satisfies h ≤ Φ(h) the sequence (Φkh)k is monotone increasing f =

k Φk(h) exists in X and satisfies f ≤ Φ(f )

f takes finite values f is (the unique) fixed point of Φ

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-41
SLIDE 41

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Finite values

Theorem 1 For this type of Φ: f takes finite values.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-42
SLIDE 42

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Finite values

Theorem 1 For this type of Φ: f takes finite values. Sketch of Proof: if f takes infinite values on Y , let U = {u ∈ Y | f (u) = ∞} Since U = ∅ there exists a minimal element y in (U, ≺). In particular f (y) = ∞. By finiteness of h(y) we have y is not minimal in Y . f (y) ≤ Φf (y) so we have Φf (y) = ∞. But Φf (y) = h(y) + Ψy((f (u))u∈Yy ). In view of the minimality of y in U we have f (u) < ∞ for u ∈ Yy, so by the finiteness of Ψy a contradiction follows.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-43
SLIDE 43

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Existence of a fixed point

Theorem 2: For this type of Φ: f is a fixed point of Φ.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-44
SLIDE 44

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Existence of a fixed point

Theorem 2: For this type of Φ: f is a fixed point of Φ. Sketch of Proof: We show that for fixed y ∈ Y and ε > 0 arbitrary, Φf (y) ≤ f (y) + ε. The only non trivial case is y not minimal.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-45
SLIDE 45

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Existence of a fixed point

Theorem 2: For this type of Φ: f is a fixed point of Φ. Sketch of Proof: We show that for fixed y ∈ Y and ε > 0 arbitrary, Φf (y) ≤ f (y) + ε. The only non trivial case is y not minimal. Apply the limit condition on Ψy to obtain δ > 0 such that Ψy(s) ≤ ε whenever s ≤ δ.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-46
SLIDE 46

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Existence of a fixed point

Theorem 2: For this type of Φ: f is a fixed point of Φ. Sketch of Proof: We show that for fixed y ∈ Y and ε > 0 arbitrary, Φf (y) ≤ f (y) + ε. The only non trivial case is y not minimal. Apply the limit condition on Ψy to obtain δ > 0 such that Ψy(s) ≤ ε whenever s ≤ δ. We use the pointwise expression λ(Φnh)n(f ) = supu∈Y limsupn→∞p(f (u), Φnh(u))

  • f the complexity approach space.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-47
SLIDE 47

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

For u ∈ Yy we take ju such that for k ≥ ju p(f (u), Φkh(u))) ≤ δ f (u).(f (u) − δ) · · · f (u) ≤ Φkh(u)) + δ whenever k ≥ ju

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-48
SLIDE 48

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

For u ∈ Yy we take ju such that for k ≥ ju p(f (u), Φkh(u))) ≤ δ f (u).(f (u) − δ) · · · f (u) ≤ Φkh(u)) + δ whenever k ≥ ju Fix k as the maximum of the indices in the finite set {ju | u ∈ Yy}. Φf (y) = h(y) + Ψy((f (u))u∈Yy )

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-49
SLIDE 49

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

For u ∈ Yy we take ju such that for k ≥ ju p(f (u), Φkh(u))) ≤ δ f (u).(f (u) − δ) · · · f (u) ≤ Φkh(u)) + δ whenever k ≥ ju Fix k as the maximum of the indices in the finite set {ju | u ∈ Yy}. Φf (y) = h(y) + Ψy((f (u))u∈Yy ) Φf (y) ≤ h(y) + Ψy(δ + (Φkh(u))u∈Yy )

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-50
SLIDE 50

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

For u ∈ Yy we take ju such that for k ≥ ju p(f (u), Φkh(u))) ≤ δ f (u).(f (u) − δ) · · · f (u) ≤ Φkh(u)) + δ whenever k ≥ ju Fix k as the maximum of the indices in the finite set {ju | u ∈ Yy}. Φf (y) = h(y) + Ψy((f (u))u∈Yy ) Φf (y) ≤ h(y) + Ψy(δ + (Φkh(u))u∈Yy ) Φf (y) ≤ h(y) + Ψy(δ) + Ψy((Φkh(u))u∈Yy )

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-51
SLIDE 51

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

For u ∈ Yy we take ju such that for k ≥ ju p(f (u), Φkh(u))) ≤ δ f (u).(f (u) − δ) · · · f (u) ≤ Φkh(u)) + δ whenever k ≥ ju Fix k as the maximum of the indices in the finite set {ju | u ∈ Yy}. Φf (y) = h(y) + Ψy((f (u))u∈Yy ) Φf (y) ≤ h(y) + Ψy(δ + (Φkh(u))u∈Yy ) Φf (y) ≤ h(y) + Ψy(δ) + Ψy((Φkh(u))u∈Yy ) ≤ Ψy(δ) + Φk+1h(y) ≤ Ψy(δ) + f (y) Φf (y) ≤ ε + f (y).

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-52
SLIDE 52

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Examples 1

All the examples Mergesort, Quicksort, · · · fit into the framework

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-53
SLIDE 53

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Examples 1

All the examples Mergesort, Quicksort, · · · fit into the framework Several variables as inputs (ExpoDC algorithm): Y = N0 × N0 and let Z = [0, ∞]. Φ on X = Z Y : for g ∈ X Φg(m, n) =      if n = 1 g(m, n

2) + M( mn 2 , mn 2 )

if n is even g(m, n − 1) + M(m, (n − 1)m)

  • therwise

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-54
SLIDE 54

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Examples 1

All the examples Mergesort, Quicksort, · · · fit into the framework Several variables as inputs (ExpoDC algorithm): Y = N0 × N0 and let Z = [0, ∞]. Φ on X = Z Y : for g ∈ X Φg(m, n) =      if n = 1 g(m, n

2) + M( mn 2 , mn 2 )

if n is even g(m, n − 1) + M(m, (n − 1)m)

  • therwise

Binary relation on Y : (m, n) ≺ (m′, n′) ⇔ m = m′ and n < n′.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-55
SLIDE 55

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

For y = (m, n) not minimal, Ψy : [0, ∞]Yy → [0, ∞] is defined by Ψy(a) =

  • am, n

2

if n is even am,n−1

  • therwise

with a = (am,1, · · · , am,n−1).

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-56
SLIDE 56

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Examples 2

The vertex covering problem Given a graph G = (V , E), does there exist a subset W ⊆ V

  • f the set of vertices of G, of size k, such that for every edge

(u, v) ∈ E of G, either u ∈ W or v ∈ W ?

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-57
SLIDE 57

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Examples 2

The vertex covering problem Given a graph G = (V , E), does there exist a subset W ⊆ V

  • f the set of vertices of G, of size k, such that for every edge

(u, v) ∈ E of G, either u ∈ W or v ∈ W ? Inputs (n, k), where n is the number of all vertices of the graph and k is the size of the subset W : Take Y = {(n, k) ∈ N × N | k ≤ n} endowed with the strict pointwise relation, Z = [0, ∞] and h = pr1 : Y → Z the first projection.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-58
SLIDE 58

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Φ : Z Y → Z Y is Φ(g)(n, k) =

  • h(n, k) + 2g(n − 1, k − 1)

(n, k) not minimal h(n, k)

  • therwise.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-59
SLIDE 59

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Φ : Z Y → Z Y is Φ(g)(n, k) =

  • h(n, k) + 2g(n − 1, k − 1)

(n, k) not minimal h(n, k)

  • therwise.

For (n, k) not minimal Ψ(n,k) : Z Y(n,k) → Z is defined as Ψ(n,k)(a) = 2a(n−1,k−1).

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-60
SLIDE 60

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Upper- and lowerbounds

If Φ is of the same type and g ∈ X is such that Φ(g) ≤ g (resp g ≤ Φg) then the fixed point f satisfies f ≤ g (resp g ≤ f ).

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-61
SLIDE 61

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Upper- and lowerbounds

If Φ is of the same type and g ∈ X is such that Φ(g) ≤ g (resp g ≤ Φg) then the fixed point f satisfies f ≤ g (resp g ≤ f ). For y not minimal. Suppose f (u) ≤ g(u) for all u ∈ Yy. The result follows using monotonicity of Ψy and the fact that f (y) = Φf (y) = h(y) + Ψy((f (u))u∈Yy ) ≤ h(y) + Ψy((g(u))u∈Yy ) = Φg(y) ≤ g(y).

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-62
SLIDE 62

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Comparison

Z = (]0, ∞], ≤) with p(x, y) = ( 1

y − 1 x ) ∨ 0.

For Y = N and X = (]0, ∞]N, ≤) and (gk)k and f in X.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-63
SLIDE 63

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Comparison

Z = (]0, ∞], ≤) with p(x, y) = ( 1

y − 1 x ) ∨ 0.

For Y = N and X = (]0, ∞]N, ≤) and (gk)k and f in X. Complexity quasi metric structure: dC(f , gk) = Σn∈N 1 2n .[( 1 gk(n) − 1 f (n)) ∨ 0] = Σn∈N 1 2n .p(f (n), gk(n))

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-64
SLIDE 64

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

Comparison

Z = (]0, ∞], ≤) with p(x, y) = ( 1

y − 1 x ) ∨ 0.

For Y = N and X = (]0, ∞]N, ≤) and (gk)k and f in X. Complexity quasi metric structure: dC(f , gk) = Σn∈N 1 2n .[( 1 gk(n) − 1 f (n)) ∨ 0] = Σn∈N 1 2n .p(f (n), gk(n)) (C, dC) is not compatible with the trace of the producttopology.

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-65
SLIDE 65

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

complexity approach structure: Categorical product on X =]0, ∞]N The gauge on X is the saturation of the ideal generated by the collection {p ◦ prn × prn | n ∈ N} where p ◦ prn × prn(f , g) = (

1 g(n) − 1 f (n)) ∨ 0

Eva Colebunders Products in the category of approach spaces as models for complexit

slide-66
SLIDE 66

Complexity quasi metric spaces Changing the categorical setting Approach complexity space Fixed points Comparison to the complexit

complexity approach structure: Categorical product on X =]0, ∞]N The gauge on X is the saturation of the ideal generated by the collection {p ◦ prn × prn | n ∈ N} where p ◦ prn × prn(f , g) = (

1 g(n) − 1 f (n)) ∨ 0

λ(gk)k(f ) = supn∈Nlimsupk→∞( 1 gk(n)− 1 f (n))∨0 λ(gk)k(f ) ≤ α ⇔ ∀η > 0, ∀n ∈ N, ∃jn, ∀k ≥ jn ( 1 gk(n)− 1 f (n))∨0 < α+η

Eva Colebunders Products in the category of approach spaces as models for complexit