+ All Categories
Home > Documents > Order-Degree Curves for Hypergeometric Creative … Telescoping INPUT: something like f(n;k) := ( 1)...

Order-Degree Curves for Hypergeometric Creative … Telescoping INPUT: something like f(n;k) := ( 1)...

Date post: 21-Mar-2018
Category:
Upload: lamduong
View: 225 times
Download: 3 times
Share this document with a friend
75
Order-Degree Curves for Hypergeometric Creative Telescoping Shaoshi Chen NCSU, Raleigh, USA Manuel Kauers RISC, Linz, Austria 1
Transcript

Order-Degree Curves for

Hypergeometric Creative

Telescoping

Shaoshi ChenNCSU, Raleigh, USA

Manuel KauersRISC, Linz, Austria

1

Creative Telescoping

2

Creative Telescoping

INPUT: something like f(n, k) := (−1)k 1k+1

(nk

)(2n+k

k

)

2

Creative Telescoping

INPUT: something like f(n, k) := (−1)k 1k+1

(nk

)(2n+k

k

)OUTPUT: something like

(4n+ 2)f(n, k) + (n+ 2)f(n+ 1, k) = g(n, k + 1)− g(n, k)

where g(n, k) = k(k+1)(2−2k+9n−kn+10n2)2n(1−k+n)(2n+1) f(n, k).

2

Creative Telescoping

INPUT: something like f(n, k) := (−1)k 1k+1

(nk

)(2n+k

k

)OUTPUT: something like

(4n+ 2)f(n, k) + (n+ 2)f(n+ 1, k) = g(n, k + 1)− g(n, k)

where g(n, k) = k(k+1)(2−2k+9n−kn+10n2)2n(1−k+n)(2n+1) f(n, k).

2

Creative Telescoping

INPUT: something like f(n, k) := (−1)k 1k+1

(nk

)(2n+k

k

)OUTPUT: something like

(4n+ 2)f(n, k) + (n+ 2)f(n+ 1, k) = g(n, k + 1)− g(n, k)

where g(n, k) = k(k+1)(2−2k+9n−kn+10n2)2n(1−k+n)(2n+1) f(n, k).

polynomials in n only

2

Creative Telescoping

INPUT: something like f(n, k) := (−1)k 1k+1

(nk

)(2n+k

k

)OUTPUT: something like

(4n+ 2)f(n, k) + (n+ 2)f(n+ 1, k) = g(n, k + 1)− g(n, k)

where g(n, k) = k(k+1)(2−2k+9n−kn+10n2)2n(1−k+n)(2n+1) f(n, k).

polynomials in n only

shift(s) in n only

2

Creative Telescoping

INPUT: something like f(n, k) := (−1)k 1k+1

(nk

)(2n+k

k

)OUTPUT: something like

(4n+ 2)f(n, k) + (n+ 2)f(n+ 1, k) = g(n, k + 1)− g(n, k)

where g(n, k) = k(k+1)(2−2k+9n−kn+10n2)2n(1−k+n)(2n+1) f(n, k).

polynomials in n only

shift(s) in n only︷ ︸︸ ︷= ∆kg(n, k)

2

Creative Telescoping

INPUT: something like f(n, k) := (−1)k 1k+1

(nk

)(2n+k

k

)OUTPUT: something like

(4n+ 2)f(n, k) + (n+ 2)f(n+ 1, k) = g(n, k + 1)− g(n, k)

where g(n, k) = k(k+1)(2−2k+9n−kn+10n2)2n(1−k+n)(2n+1) f(n, k).

polynomials in n only

shift(s) in n only︷ ︸︸ ︷= Σ−1

k g(n, k)

2

Creative Telescoping

INPUT: something like f(n, k) := (−1)k 1k+1

(nk

)(2n+k

k

)OUTPUT: something like

(4n+ 2)f(n, k) + (n+ 2)f(n+ 1, k) = g(n, k + 1)− g(n, k)

where g(n, k) = k(k+1)(2−2k+9n−kn+10n2)2n(1−k+n)(2n+1) f(n, k).

polynomials in n only

shift(s) in n only︷ ︸︸ ︷= Σ−1

k g(n, k)

rational function in n and k

2

Creative Telescoping

INPUT: a hypergeometric term f(n, k)

2

Creative Telescoping

INPUT: a hypergeometric term f(n, k)

i.e., f(n+1,k)f(n,k) ∈ K(n, k) and f(n,k+1)

f(n,k) ∈ K(n, k)

2

Creative Telescoping

INPUT: a hypergeometric term f(n, k)

OUTPUT: T ∈ K[n, Sn] \ {0} and Q ∈ K(n, k) such that

T · f(n, k) = (Sk − 1) ·Qf(n, k)

∣∣∣ ∑k

2

Creative Telescoping

INPUT: a hypergeometric term f(n, k)

OUTPUT: T ∈ K[n, Sn] \ {0} and Q ∈ K(n, k) such that

T · f(n, k) = (Sk − 1) ·Qf(n, k)

∣∣∣ ∑k

“telescoper”

2

Creative Telescoping

INPUT: a hypergeometric term f(n, k)

OUTPUT: T ∈ K[n, Sn] \ {0} and Q ∈ K(n, k) such that

T · f(n, k) = (Sk − 1) ·Qf(n, k)

∣∣∣ ∑k

“telescoper” “certificate”

2

Creative Telescoping

INPUT: a hypergeometric term f(n, k)

OUTPUT: T ∈ K[n, Sn] \ {0} and Q ∈ K(n, k) such that

T · f(n, k) = (Sk − 1) ·Qf(n, k)

∣∣∣ ∑k

2

Creative Telescoping

INPUT: a hypergeometric term f(n, k)

OUTPUT: T ∈ K[n, Sn] \ {0} and Q ∈ K(n, k) such that

T · f(n, k) = (Sk − 1) ·Qf(n, k)∣∣∣ ∑k

2

Creative Telescoping

INPUT: a hypergeometric term f(n, k)

OUTPUT: T ∈ K[n, Sn] \ {0} and Q ∈ K(n, k) such that∑k

T · f(n, k) =∑k

(Sk − 1) ·Qf(n, k)

2

Creative Telescoping

INPUT: a hypergeometric term f(n, k)

OUTPUT: T ∈ K[n, Sn] \ {0} and Q ∈ K(n, k) such that

T ·∑k

f(n, k) =∑k

(Sk − 1) ·Qf(n, k)

2

Creative Telescoping

INPUT: a hypergeometric term f(n, k)

OUTPUT: T ∈ K[n, Sn] \ {0} and Q ∈ K(n, k) such that

T ·∑k

f(n, k) = 0 (usually)

2

Focus on the telescoper

T =

degree d︷ ︸︸ ︷

(a0,0 + a0,1n+ a0,2n

2 + · · ·+ a0,dnd)

+(a1,0 + a1,1n+ a1,2n

2 + · · ·+ a1,dnd)Sn

+(a2,0 + a2,1n+ a2,2n

2 + · · ·+ a2,dnd)S2n

+ . . .

+(ar,0 + ar,1n+ ar,2n

2 + · · ·+ ar,dnd)Srn

order r

3

Focus on the telescoper

T =

degree d︷ ︸︸ ︷

(a0,0 + a0,1n+ a0,2n

2 + · · ·+ a0,dnd)

+(a1,0 + a1,1n+ a1,2n

2 + · · ·+ a1,dnd)Sn

+(a2,0 + a2,1n+ a2,2n

2 + · · ·+ a2,dnd)S2n

+ . . .

+(ar,0 + ar,1n+ ar,2n

2 + · · ·+ ar,dnd)Srn

order r

3

Focus on the telescoper

T =

degree d︷ ︸︸ ︷(a0,0 + a0,1n+ a0,2n

2 + · · ·+ a0,dnd)

+(a1,0 + a1,1n+ a1,2n

2 + · · ·+ a1,dnd)Sn

+(a2,0 + a2,1n+ a2,2n

2 + · · ·+ a2,dnd)S2n

+ . . .

+(ar,0 + ar,1n+ ar,2n

2 + · · ·+ ar,dnd)Srn

order r

3

Focus on the telescoper

Question: For a given hypergeometric term f(n, k), what are theorder r and the degree d of the corresponding telescoper?

Answer: This is not a good question. “The” telescoper is notuniquely determined by f(n, k)!

Instead, the set of all telescopers for a fixed term f(n, k) forms aleft ideal in the operator algebra K[n, Sn].

4

Focus on the telescoper

Question: For a given hypergeometric term f(n, k), what are theorder r and the degree d of the corresponding telescoper?

Answer: This is not a good question. “The” telescoper is notuniquely determined by f(n, k)!

Instead, the set of all telescopers for a fixed term f(n, k) forms aleft ideal in the operator algebra K[n, Sn].

4

Focus on the telescoper

Question: For a given hypergeometric term f(n, k), what are theorder r and the degree d of the corresponding telescoper?

Answer: This is not a good question. “The” telescoper is notuniquely determined by f(n, k)!

Instead, the set of all telescopers for a fixed term f(n, k) forms aleft ideal in the operator algebra K[n, Sn].

4

The Order-Degree-Curve

order

degree

5

The Order-Degree-Curve

T

order

degree

5

The Order-Degree-Curve

T

︸ ︷︷ ︸r

order

degree

5

The Order-Degree-Curve

T

︸ ︷︷ ︸r

d

order

degree

5

The Order-Degree-Curve

order

degree

5

The Order-Degree-Curve

order

degree

5

The Order-Degree-Curve

order

degree

5

The Order-Degree-Curve

order

degree

5

The Order-Degree-Curve

order

degree

5

The Order-Degree-Curve

order

degree

5

The Order-Degree-Curve

Theorem.

I Consider a proper hypergeometric term

f(n, k) = pol(n, k)xnykM∏

m=1

Γ(amn+a′mk+a′′m)Γ(bmn−b′mk+b′′m)Γ(umn+u′

mk+u′′m)Γ(vmn−v′mk+v′′m) .

I There exists a telescoper of order r and degree d whenever

d >Ar +B

r + C

whereI A = ϑν − 1, B = 2 deg pol + |µ|+ 3− (1 + |µ|)ν, C = 1− ν.I µ =

∑Mm=1(am + bm − um − vm)

I ν = max{∑M

m=1(a′m + v′m),∑M

m=1(u′m + b′m)}

I ϑ = max{∑M

m=1(am + bm),∑M

m=1(um + vm)}

6

The Order-Degree-Curve

Theorem.

I Consider a proper hypergeometric term

f(n, k) = pol(n, k)xnykM∏

m=1

Γ(amn+a′mk+a′′m)Γ(bmn−b′mk+b′′m)Γ(umn+u′

mk+u′′m)Γ(vmn−v′mk+v′′m) .

I There exists a telescoper of order r and degree d whenever

d >Ar +B

r + C

whereI A = ϑν − 1, B = 2 deg pol + |µ|+ 3− (1 + |µ|)ν, C = 1− ν.I µ =

∑Mm=1(am + bm − um − vm)

I ν = max{∑M

m=1(a′m + v′m),∑M

m=1(u′m + b′m)}

I ϑ = max{∑M

m=1(am + bm),∑M

m=1(um + vm)}

6

The Order-Degree-Curve

Theorem.

I Consider a proper hypergeometric term

f(n, k) = pol(n, k)xnykM∏

m=1

Γ(amn+a′mk+a′′m)Γ(bmn−b′mk+b′′m)Γ(umn+u′

mk+u′′m)Γ(vmn−v′mk+v′′m) .

I There exists a telescoper of order r and degree d whenever

d >Ar +B

r + C

whereI A = ϑν − 1, B = 2 deg pol + |µ|+ 3− (1 + |µ|)ν, C = 1− ν.I µ =

∑Mm=1(am + bm − um − vm)

I ν = max{∑M

m=1(a′m + v′m),∑M

m=1(u′m + b′m)}

I ϑ = max{∑M

m=1(am + bm),∑M

m=1(um + vm)}

6

The Order-Degree-Curve

Theorem.

I Consider a proper hypergeometric term

f(n, k) = pol(n, k)xnykM∏

m=1

Γ(amn+a′mk+a′′m)Γ(bmn−b′mk+b′′m)Γ(umn+u′

mk+u′′m)Γ(vmn−v′mk+v′′m) .

I There exists a telescoper of order r and degree d whenever

d >Ar +B

r + C

where

I A = ϑν − 1, B = 2 deg pol + |µ|+ 3− (1 + |µ|)ν, C = 1− ν.I µ =

∑Mm=1(am + bm − um − vm)

I ν = max{∑M

m=1(a′m + v′m),∑M

m=1(u′m + b′m)}

I ϑ = max{∑M

m=1(am + bm),∑M

m=1(um + vm)}

6

The Order-Degree-Curve

Theorem.

I Consider a proper hypergeometric term

f(n, k) = pol(n, k)xnykM∏

m=1

Γ(amn+a′mk+a′′m)Γ(bmn−b′mk+b′′m)Γ(umn+u′

mk+u′′m)Γ(vmn−v′mk+v′′m) .

I There exists a telescoper of order r and degree d whenever

d >Ar +B

r + C

whereI A = ϑν − 1, B = 2 deg pol + |µ|+ 3− (1 + |µ|)ν, C = 1− ν.

I µ =∑M

m=1(am + bm − um − vm)

I ν = max{∑M

m=1(a′m + v′m),∑M

m=1(u′m + b′m)}

I ϑ = max{∑M

m=1(am + bm),∑M

m=1(um + vm)}

6

The Order-Degree-Curve

Theorem.

I Consider a proper hypergeometric term

f(n, k) = pol(n, k)xnykM∏

m=1

Γ(amn+a′mk+a′′m)Γ(bmn−b′mk+b′′m)Γ(umn+u′

mk+u′′m)Γ(vmn−v′mk+v′′m) .

I There exists a telescoper of order r and degree d whenever

d >Ar +B

r + C

whereI A = ϑν − 1, B = 2 deg pol + |µ|+ 3− (1 + |µ|)ν, C = 1− ν.I µ =

∑Mm=1(am + bm − um − vm)

I ν = max{∑M

m=1(a′m + v′m),∑M

m=1(u′m + b′m)}

I ϑ = max{∑M

m=1(am + bm),∑M

m=1(um + vm)}

6

The Order-Degree-Curve

Theorem.

I Consider a proper hypergeometric term

f(n, k) = pol(n, k)xnykM∏

m=1

Γ(amn+a′mk+a′′m)Γ(bmn−b′mk+b′′m)Γ(umn+u′

mk+u′′m)Γ(vmn−v′mk+v′′m) .

I There exists a telescoper of order r and degree d whenever

d >Ar +B

r + C

whereI A = ϑν − 1, B = 2 deg pol + |µ|+ 3− (1 + |µ|)ν, C = 1− ν.I µ =

∑Mm=1(am + bm − um − vm)

I ν = max{∑M

m=1(a′m + v′m),∑M

m=1(u′m + b′m)}

I ϑ = max{∑M

m=1(am + bm),∑M

m=1(um + vm)}

6

The Order-Degree-Curve

Theorem.

I Consider a proper hypergeometric term

f(n, k) = pol(n, k)xnykM∏

m=1

Γ(amn+a′mk+a′′m)Γ(bmn−b′mk+b′′m)Γ(umn+u′

mk+u′′m)Γ(vmn−v′mk+v′′m) .

I There exists a telescoper of order r and degree d whenever

d >Ar +B

r + C

whereI A = ϑν − 1, B = 2 deg pol + |µ|+ 3− (1 + |µ|)ν, C = 1− ν.I µ =

∑Mm=1(am + bm − um − vm)

I ν = max{∑M

m=1(a′m + v′m),∑M

m=1(u′m + b′m)}

I ϑ = max{∑M

m=1(am + bm),∑M

m=1(um + vm)}

6

The Order-Degree-Curve

Example 1: (n2+k2+1)Γ(2n+3k)Γ(2n−k)

d >7r + 5

r − 3

Example 2: Γ(2n+k)Γ(n−k+2)Γ(2n−k)Γ(n+2k)

d >8r − 1

r − 2

7

The Order-Degree-Curve

Example 1: (n2+k2+1)Γ(2n+3k)Γ(2n−k)

d >7r + 5

r − 3

Example 2: Γ(2n+k)Γ(n−k+2)Γ(2n−k)Γ(n+2k)

d >8r − 1

r − 2

7

The Order-Degree-Curve

Example 1: (n2+k2+1)Γ(2n+3k)Γ(2n−k)

d >7r + 5

r − 3

5 10 15

10

20

30

Example 2: Γ(2n+k)Γ(n−k+2)Γ(2n−k)Γ(n+2k)

d >8r − 1

r − 2

5 10 15

10

20

30

7

The Order-Degree-Curve

Example 1: (n2+k2+1)Γ(2n+3k)Γ(2n−k)

d >7r + 5

r − 3

5 10 15

10

20

30

Example 2: Γ(2n+k)Γ(n−k+2)Γ(2n−k)Γ(n+2k)

d >8r − 1

r − 2

5 10 15

10

20

30

7

The Order-Degree-Curve

Example 1: (n2+k2+1)Γ(2n+3k)Γ(2n−k)

d >7r + 5

r − 3

5 10 15

10

20

30

sometimes tight

Example 2: Γ(2n+k)Γ(n−k+2)Γ(2n−k)Γ(n+2k)

d >8r − 1

r − 2

5 10 15

10

20

30

sometimes not

7

The Order-Degree-Curve

The curve overshoots much when f(n, k) is a rational function.

Therefore we give a separate formula for this case in the paper.

Also the formula for this special case has the form d > Ar+Br−C .

The coefficients A,B,C now depend on quantities appearing in acertain representation of f(n, k) as computed in Le’s algorithm.

The refined formula may or may not be tight, but at least it seemsto overshoot far less than the general formula.

See the paper for details and examples.

8

The Order-Degree-Curve

The curve overshoots much when f(n, k) is a rational function.

Therefore we give a separate formula for this case in the paper.

Also the formula for this special case has the form d > Ar+Br−C .

The coefficients A,B,C now depend on quantities appearing in acertain representation of f(n, k) as computed in Le’s algorithm.

The refined formula may or may not be tight, but at least it seemsto overshoot far less than the general formula.

See the paper for details and examples.

8

The Order-Degree-Curve

The curve overshoots much when f(n, k) is a rational function.

Therefore we give a separate formula for this case in the paper.

Also the formula for this special case has the form d > Ar+Br−C .

The coefficients A,B,C now depend on quantities appearing in acertain representation of f(n, k) as computed in Le’s algorithm.

The refined formula may or may not be tight, but at least it seemsto overshoot far less than the general formula.

See the paper for details and examples.

8

The Order-Degree-Curve

The curve overshoots much when f(n, k) is a rational function.

Therefore we give a separate formula for this case in the paper.

Also the formula for this special case has the form d > Ar+Br−C .

The coefficients A,B,C now depend on quantities appearing in acertain representation of f(n, k) as computed in Le’s algorithm.

The refined formula may or may not be tight, but at least it seemsto overshoot far less than the general formula.

See the paper for details and examples.

8

The Order-Degree-Curve

The curve overshoots much when f(n, k) is a rational function.

Therefore we give a separate formula for this case in the paper.

Also the formula for this special case has the form d > Ar+Br−C .

The coefficients A,B,C now depend on quantities appearing in acertain representation of f(n, k) as computed in Le’s algorithm.

The refined formula may or may not be tight, but at least it seemsto overshoot far less than the general formula.

See the paper for details and examples.

8

The Order-Degree-Curve

The curve overshoots much when f(n, k) is a rational function.

Therefore we give a separate formula for this case in the paper.

Also the formula for this special case has the form d > Ar+Br−C .

The coefficients A,B,C now depend on quantities appearing in acertain representation of f(n, k) as computed in Le’s algorithm.

The refined formula may or may not be tight, but at least it seemsto overshoot far less than the general formula.

See the paper for details and examples.

8

Consequences

Even if it may not be accurate, we can use the curve to estimatethe shapes of some interesting telescopers, before computing them.

order

degree

9

Consequences

Even if it may not be accurate, we can use the curve to estimatethe shapes of some interesting telescopers, before computing them.

order

degree

9

Consequences

Even if it may not be accurate, we can use the curve to estimatethe shapes of some interesting telescopers, before computing them.

order

degree

minimal order

9

Consequences

Even if it may not be accurate, we can use the curve to estimatethe shapes of some interesting telescopers, before computing them.

order

degree

minimal order

minimal degree

9

Consequences

Even if it may not be accurate, we can use the curve to estimatethe shapes of some interesting telescopers, before computing them.

order

degree

minimal order

minimal degree

minimal telescoper size

9

Consequences

Even if it may not be accurate, we can use the curve to estimatethe shapes of some interesting telescopers, before computing them.

order

degree

minimal order

minimal degree

minimal telescoper size

minimal total size

9

Consequences

Even if it may not be accurate, we can use the curve to estimatethe shapes of some interesting telescopers, before computing them.

order

degree

minimal order

minimal degree

minimal cost

minimal telescoper size

minimal total size

9

Consequences

I For currently feasible input sizes, the minimal cost telescoperagrees with minimal order telescoper.

I We expect that the separation becomes measurable within thecoming few years.

I For asymptotically large input size, the difference is significant.

For τ ≥ max{ϑ, ν} and any fixed constant α > 1 we have:I O∼(τ9). . . cost for telescoper of expected minimal order rmin

I O∼(τ8). . . cost for telescoper of order α rmin.

I Under appropriate assumptions, the optimal choice of α turnsout to be 1.2.

I Similar effects happen in other summation/integrationcontexts (cf. e.g. Alin’s talk on Wednesday 12:05–12:30).

10

Consequences

I For currently feasible input sizes, the minimal cost telescoperagrees with minimal order telescoper.

I We expect that the separation becomes measurable within thecoming few years.

I For asymptotically large input size, the difference is significant.

For τ ≥ max{ϑ, ν} and any fixed constant α > 1 we have:I O∼(τ9). . . cost for telescoper of expected minimal order rmin

I O∼(τ8). . . cost for telescoper of order α rmin.

I Under appropriate assumptions, the optimal choice of α turnsout to be 1.2.

I Similar effects happen in other summation/integrationcontexts (cf. e.g. Alin’s talk on Wednesday 12:05–12:30).

10

Consequences

I For currently feasible input sizes, the minimal cost telescoperagrees with minimal order telescoper.

I We expect that the separation becomes measurable within thecoming few years.

I For asymptotically large input size, the difference is significant.

For τ ≥ max{ϑ, ν} and any fixed constant α > 1 we have:I O∼(τ9). . . cost for telescoper of expected minimal order rmin

I O∼(τ8). . . cost for telescoper of order α rmin.

I Under appropriate assumptions, the optimal choice of α turnsout to be 1.2.

I Similar effects happen in other summation/integrationcontexts (cf. e.g. Alin’s talk on Wednesday 12:05–12:30).

10

Consequences

I For currently feasible input sizes, the minimal cost telescoperagrees with minimal order telescoper.

I We expect that the separation becomes measurable within thecoming few years.

I For asymptotically large input size, the difference is significant.For τ ≥ max{ϑ, ν} and any fixed constant α > 1 we have:

I O∼(τ9). . . cost for telescoper of expected minimal order rmin

I O∼(τ8). . . cost for telescoper of order α rmin.

I Under appropriate assumptions, the optimal choice of α turnsout to be 1.2.

I Similar effects happen in other summation/integrationcontexts (cf. e.g. Alin’s talk on Wednesday 12:05–12:30).

10

Consequences

I For currently feasible input sizes, the minimal cost telescoperagrees with minimal order telescoper.

I We expect that the separation becomes measurable within thecoming few years.

I For asymptotically large input size, the difference is significant.For τ ≥ max{ϑ, ν} and any fixed constant α > 1 we have:

I O∼(τ9). . . cost for telescoper of expected minimal order rmin

I O∼(τ8). . . cost for telescoper of order α rmin.

I Under appropriate assumptions, the optimal choice of α turnsout to be 1.2.

I Similar effects happen in other summation/integrationcontexts (cf. e.g. Alin’s talk on Wednesday 12:05–12:30).

10

Consequences

I For currently feasible input sizes, the minimal cost telescoperagrees with minimal order telescoper.

I We expect that the separation becomes measurable within thecoming few years.

I For asymptotically large input size, the difference is significant.For τ ≥ max{ϑ, ν} and any fixed constant α > 1 we have:

I O∼(τ9). . . cost for telescoper of expected minimal order rmin

I O∼(τ8). . . cost for telescoper of order α rmin.

I Under appropriate assumptions, the optimal choice of α turnsout to be 1.2.

I Similar effects happen in other summation/integrationcontexts (cf. e.g. Alin’s talk on Wednesday 12:05–12:30).

10

Consequences

I For currently feasible input sizes, the minimal cost telescoperagrees with minimal order telescoper.

I We expect that the separation becomes measurable within thecoming few years.

I For asymptotically large input size, the difference is significant.For τ ≥ max{ϑ, ν} and any fixed constant α > 1 we have:

I O∼(τ9). . . cost for telescoper of expected minimal order rmin

I O∼(τ8). . . cost for telescoper of order α rmin.

I Under appropriate assumptions, the optimal choice of α turnsout to be 1.2.

I Similar effects happen in other summation/integrationcontexts (cf. e.g. Alin’s talk on Wednesday 12:05–12:30).

10

Open Questions

I What is the smallest problem size for which it pays off tocompute a non-minimal telescoper?

I What is the “true curve” which (generically) does notovershoot? Is it also a hyperbola?

I What is the deeper reason behind all these order/degreephenomena discovered recently?

I What is the right question to be asked in the case of severalvariables?

11

Open Questions

I What is the smallest problem size for which it pays off tocompute a non-minimal telescoper?

I What is the “true curve” which (generically) does notovershoot? Is it also a hyperbola?

I What is the deeper reason behind all these order/degreephenomena discovered recently?

I What is the right question to be asked in the case of severalvariables?

11

Open Questions

I What is the smallest problem size for which it pays off tocompute a non-minimal telescoper?

I What is the “true curve” which (generically) does notovershoot? Is it also a hyperbola?

I What is the deeper reason behind all these order/degreephenomena discovered recently?

I What is the right question to be asked in the case of severalvariables?

11

Open Questions

I What is the smallest problem size for which it pays off tocompute a non-minimal telescoper?

I What is the “true curve” which (generically) does notovershoot? Is it also a hyperbola?

I What is the deeper reason behind all these order/degreephenomena discovered recently?

I What is the right question to be asked in the case of severalvariables?

11

Open Questions

I What is the smallest problem size for which it pays off tocompute a non-minimal telescoper?

I What is the “true curve” which (generically) does notovershoot? Is it also a hyperbola?

I What is the deeper reason behind all these order/degreephenomena discovered recently?

I What is the right question to be asked in the case of severalvariables?

11

12


Recommended