43
Linear algebra 1 Linear operators and representations Motivating example: Web start-up Vector space and basis Eigenvector-eigenvalue analysis + ^ [ 1 β€² 2 β€² ] = [ 1 ,1 1 , 2 2 ,1 2 , 2 ][ 1 2 ] ^ =

Linear algebra

  • Upload
    rowdy

  • View
    33

  • Download
    0

Embed Size (px)

DESCRIPTION

Linear algebra. Motivating example: Web start-up. Eigenvector-eigenvalue analysis. Vector space and basis. Linear operators and representations. +. Example: Modeling a freemium cloud data storage business. Free. Premium. +. +. +. - PowerPoint PPT Presentation

Citation preview

Page 1: Linear algebra

Linear algebra

1

Linear operators and representations

Motivating example: Web start-up

Vector space and basis

Eigenvector-eigenvalue analysis

+

οΏ½Μ‚οΏ½ �⃑�

𝑣

[𝑣1′𝑣2β€² ]=[𝔸 1 ,1 𝔸 1 ,2

𝔸 2 ,1 𝔸 2 ,2] [𝑣1𝑣2]

οΏ½Μ‚οΏ½ �⃑�=πœ†π‘£

Page 2: Linear algebra

Example: Modeling a freemium cloud data storage business

2

Free Premium

π‘₯𝐹 π‘₯𝑃Event β€œCausal” subpopulation Fraction thereof

Recruit Premium users +1 0

+ +

+

Page 3: Linear algebra

Example: Modeling a freemium cloud data storage business

3

Free Premium

π‘₯𝐹 π‘₯𝑃Event β€œCausal” subpopulation Fraction thereof

Recruit Premium users +1 0

Upgrade Free users -1 +1

Page 4: Linear algebra

Example: Modeling a freemium cloud data storage business

4

Free Premium

π‘₯𝐹 π‘₯𝑃Event β€œCausal” subpopulation Fraction thereof

Recruit Premium users +1 0

Upgrade Free users -1 +1

Downgrade Premium users +1 -1

Page 5: Linear algebra

Example: Modeling a freemium cloud data storage business

5

Free Premium

π‘₯𝐹 π‘₯𝑃Event β€œCausal” subpopulation Fraction thereof

Recruit Premium users +1 0

Upgrade Free users -1 +1

Downgrade Premium users +1 -1

Attrition Free users -1 0

Page 6: Linear algebra

Example: Modeling a freemium cloud data storage business

6

Free Premium

π‘₯𝐹 π‘₯𝑃

π‘₯𝐹 (𝑑+βˆ† 𝑑 )=π‘₯𝐹 (𝑑 )+𝜌π‘₯ 𝑃 (𝑑 )βˆ’πœπ‘₯𝐹 (𝑑 )+𝛿π‘₯𝑃 (𝑑 )βˆ’π›Όπ‘₯𝐹 (𝑑 )

Event β€œCausal” subpopulation Fraction thereof

Recruit Premium users +1 0

Upgrade Free users -1 +1

Downgrade Premium users +1 -1

Attrition Free users -1 0

π‘₯𝑃 (𝑑+βˆ† 𝑑 )=π‘₯𝑃 (𝑑 )+𝜌 π‘₯𝑃 (𝑑 )+𝜐 π‘₯𝐹 (𝑑 )βˆ’π›Ώπ‘₯ 𝑃 (𝑑 )βˆ’π›Ό π‘₯𝐹 (𝑑 )

Page 7: Linear algebra

Linear algebra

7

Linear operators and representations

Motivating example: Web start-up

Vector space and basis

Eigenvector-eigenvalue analysis

+

οΏ½Μ‚οΏ½ �⃑�

𝑣

[𝑣1′𝑣2β€² ]=[𝔸 1 ,1 𝔸 1 ,2

𝔸 2 ,1 𝔸 2 ,2] [𝑣1𝑣2]

οΏ½Μ‚οΏ½ �⃑�=πœ†π‘£

Page 8: Linear algebra

8

Vector

𝑣

A vector is an arrow. The position of the head in relation to the tail is expressed in terms of a magnitude and direction.

Page 9: Linear algebra

9

A set of vectors

𝑣

�⃑��⃑�

𝑦�⃑�

Page 10: Linear algebra

10

A set of vectors

𝑣

�⃑��⃑�

𝑦�⃑�

Page 11: Linear algebra

11

𝑣

�⃑��⃑�

𝑦

𝑣+𝑀

�⃑�

β€œHead-to-tail” addition of and produced a resultant vector not belonging to our original set of vectors

A set of vectors vs. a vector spaceThis scaling (doubling length in this example) of produced 2, which belongs to our original set of vectors

A vector space is a set of vectors that is β€œclosed” under scaling and vector addition. Neither scaling nor vector addition produces a result not already included in the β€œspace.”

Page 12: Linear algebra

12

BasisA vector space is a set of vectors that are β€œclosed” under scaling and vector addition.

A set of vectors , , . . .

Linear combination: addition of vectors with scalings

Used a set of vectors to prescribe a vector space!

Page 13: Linear algebra

13

Basis set: Can’t remove any vector without changing space

A vector space is a set of vectors that are β€œclosed” under scaling and vector addition.

A set of vectors , , . . .

Linear combination: addition of vectors with scalings

Basis for V vector space V2-dimensionalN

S

EW

Page 14: Linear algebra

14

Basis: Coordinate system

Linear combination: addition of vectors with scalings

A vector space is a set of vectors that are β€œclosed” under scaling and vector addition.

A set of vectors , , . . .

Page 15: Linear algebra

15

Basis: Coordinate system

Linear combination: addition of vectors with scalings

A vector space is a set of vectors that are β€œclosed” under scaling and vector addition.

A set of vectors , , . . .

𝑣=𝑣1𝑏1+𝑣2𝑏2

𝑏1 𝑏2

Page 16: Linear algebra

16

Basis: Coordinate system

Linear combination: addition of vectors with scalings

A vector space is a set of vectors that are β€œclosed” under scaling and vector addition.

A set of vectors , , . . .

𝑣=𝑣1𝑏1+𝑣2𝑏2

𝑏1 𝑏2

�⃑�1�⃑�2 𝑣=𝑣1 �⃑�1+𝑣2 �⃑�2

Page 17: Linear algebra

Linear algebra

17

Linear operators and representations

Motivating example: Web start-up

Vector space and basis

Eigenvector-eigenvalue analysis

+

οΏ½Μ‚οΏ½ �⃑�

𝑣

[𝑣1′𝑣2β€² ]=[𝔸 1 ,1 𝔸 1 ,2

𝔸 2 ,1 𝔸 2 ,2] [𝑣1𝑣2]

οΏ½Μ‚οΏ½ �⃑�=πœ†π‘£

Page 18: Linear algebra

18

Operator

𝑣

οΏ½Μ‚οΏ½ �⃑�Given a vector, an operator outputs a vector, possibly scaled and/or rotated

A function associates objects from a domain with objects in a codomain, sometimes in terms of elementary arithmetic operations.

Page 19: Linear algebra

19

Linear operators

𝛼𝑣

�̂�𝛼 �⃑�=𝛼 οΏ½Μ‚οΏ½ �⃑�

Scaling Addition

𝑣 �⃑�𝑣+𝑀

οΏ½Μ‚οΏ½ �⃑�

𝑣

�̂�𝑀�⃑�

οΏ½Μ‚οΏ½ (𝛼�⃑�+𝛽𝑀 )=𝛼 οΏ½Μ‚οΏ½ �⃑�+𝛽 οΏ½Μ‚οΏ½ �⃑�

Page 20: Linear algebra

20

Representing linear operators

οΏ½Μ‚οΏ½ (𝛼�⃑�+𝛽𝑀 )=𝛼 οΏ½Μ‚οΏ½ �⃑�+𝛽 οΏ½Μ‚οΏ½ �⃑�

𝑣=𝑣1𝑏1+𝑣2𝑏2

𝑏1 𝑏2

¿𝑣1 �̂�𝑏1+𝑣2 �̂�𝑏2𝑣 β€²= οΏ½Μ‚οΏ½ �⃑�

𝑣 β€²=𝑣1β€² 𝑏1+𝑣2

β€² 𝑏2

¿𝑣1 [( �̂�𝑏1)1𝑏1+ ( οΏ½Μ‚οΏ½ 𝑏1 )2𝑏2 ]ΒΏ οΏ½Μ‚οΏ½ (𝑣1𝑏1+𝑣2𝑏2 )

+𝑣2 [( οΏ½Μ‚οΏ½ 𝑏2 )1𝑏1+( �̂�𝑏2 )2𝑏2 ]¿𝑣1 [𝔸1 , 1𝑏1+𝔸 2 ,1𝑏2 ]

+𝑣2 [𝔸 1, 2𝑏1+𝔸2 , 2𝑏2 ]

ΒΏ (𝑣1𝔸1 , 1+𝑣2𝔸 1, 2 )𝑏1+(𝑣1𝔸 2, 1+𝑣2𝔸 2 ,2 )𝑏2

𝑣1β€² 𝑏1+𝑣2β€² 𝑏2

𝑣1β€²=𝔸 1 ,1𝑣1+𝔸 1 ,2𝑣2𝑣2β€²=𝔸2 ,1𝑣1+𝔸2 , 2𝑣2

[𝑣1′𝑣2β€² ]=[𝔸 1 ,1 𝔸 1 ,2

𝔸 2 ,1 𝔸 2 ,2] [𝑣1𝑣2]

𝔸1 , 2

Page 21: Linear algebra

21

Representing linear operators

𝑣=𝑣1𝑏1+𝑣2𝑏2

𝑏1 𝑏2

𝑣 β€²=𝑣1β€² 𝑏1+𝑣2

β€² 𝑏2

𝑣1β€²=𝔸 1 ,1𝑣1+𝔸 1 ,2𝑣2𝑣2β€²=𝔸2 ,1𝑣1+𝔸2 , 2𝑣2

[𝑣1′𝑣2β€² ]=[𝔸 1 ,1 𝔸 1 ,2

𝔸 2 ,1 𝔸 2 ,2] [𝑣1𝑣2]

οΏ½Μ‚οΏ½ (𝛼�⃑�+𝛽𝑀 )=𝛼 οΏ½Μ‚οΏ½ �⃑�+𝛽 οΏ½Μ‚οΏ½ �⃑� 𝑣 β€²= οΏ½Μ‚οΏ½ �⃑� Abstract action on vector

Relationship between coefficients

Representation in the context of a particular basis

𝑣 β€²β†’ [𝑣1′𝑣2β€² ] 𝑣→[𝑣1𝑣2]

οΏ½Μ‚οΏ½β†’ [𝔸 1 ,1 𝔸 1 ,2

𝔸 2 ,1 𝔸 2 ,2]

β€œThe vector v-prime is represented by the column vector v-prime-sub-1, v-prime-sub-2”

β€œThe operator A is represented by the matrix A”

Page 22: Linear algebra

22

Vector transformation algorithm implies matrix multiplication

𝑣 β€²= οΏ½Μ‚οΏ½ �⃑�

𝑣 β€² β€²=οΏ½Μ‚οΏ½ 𝑣 β€²

𝑣

οΏ½Μ‚οΏ½ 𝑣 β€²β†’[𝔹1 ,1 𝔹1 , 2

𝔹2 ,1 𝔹2 , 2][ 𝔸 1 ,1𝑣1+𝔸 1 ,2𝑣2𝔸 2, 1𝑣1+𝔸 2 ,2𝑣2  ]

𝑣→[𝑣1𝑣2]οΏ½Μ‚οΏ½β†’ [𝔸 1 ,1 𝔸 1 ,2

𝔸 2 ,1 𝔸 2 ,2]𝑣1β€²=𝔸 1 ,1𝑣1+𝔸 1 ,2𝑣2𝑣2β€²=𝔸2 ,1𝑣1+𝔸2 , 2𝑣2

ΒΏ [𝔹1, 1 (𝔸 1 ,1𝑣1+𝔸1 ,2𝑣2 )+𝔹1 ,2 (𝔸 2 ,1𝑣1+𝔸 2 ,2𝑣2 )𝔹2, 1 (𝔸 1 ,1𝑣1+𝔸1 ,2𝑣2 )+𝔹2 , 2 (𝔸 2 ,1𝑣1+𝔸 2 ,2𝑣2 )]

ΒΏ [ (𝔹1 ,1𝔸1 ,1+𝔹1 ,2𝔸2 , 1 )𝑣1+ (𝔹1 , 1𝔸 1, 2+𝔹1, 2𝔸 2 ,2 )𝑣2(𝔹2 ,1𝔸 1 ,1+𝔹2 ,2𝔸2 , 1 )𝑣1+ (𝔹2 ,1𝔸 1 ,2+𝔹2 ,2𝔸2 , 2 )𝑣2]

ΒΏ [𝔹1, 1𝔸 1, 1+𝔹1, 2𝔸 2 ,1 𝔹1 , 1𝔸1 , 2+𝔹1 , 2𝔸 2 ,2

𝔹2, 1𝔸 1, 1+𝔹2 ,2𝔸 2 ,1 𝔹2 , 1𝔸1 , 2+𝔹2, 2𝔸 2 ,2][𝑣1𝑣2]οΏ½Μ‚οΏ½ �̂�𝑣→[𝔹1 ,1 𝔹1 ,2

𝔹2 ,1 𝔹2 ,2] [𝔸1 ,1 𝔸1 , 2

𝔸2 , 1 𝔸2 , 2] [𝑣1𝑣2]

Page 23: Linear algebra

Linear algebra

23

Linear operators and representations

Motivating example: Web start-up

Vector space and basis

Eigenvector-eigenvalue analysis

+

οΏ½Μ‚οΏ½ �⃑�

𝑣

[𝑣1′𝑣2β€² ]=[𝔸 1 ,1 𝔸 1 ,2

𝔸 2 ,1 𝔸 2 ,2] [𝑣1𝑣2]

οΏ½Μ‚οΏ½ �⃑�=πœ†π‘£

Page 24: Linear algebra

Example: Modeling a freemium cloud data storage business

24

Free Premium

π‘₯𝐹 π‘₯𝑃Event β€œCausal” subpopulation Fraction thereof

Recruit Premium users +1 0

+ +

+

Page 25: Linear algebra

Example: Modeling a freemium cloud data storage business

25

Free Premium

π‘₯𝐹 π‘₯𝑃Event β€œCausal” subpopulation Fraction thereof

Recruit Premium users +1 0

Upgrade Free users -1 +1

Page 26: Linear algebra

Example: Modeling a freemium cloud data storage business

26

Free Premium

π‘₯𝐹 π‘₯𝑃Event β€œCausal” subpopulation Fraction thereof

Recruit Premium users +1 0

Upgrade Free users -1 +1

Downgrade Premium users +1 -1

Page 27: Linear algebra

Example: Modeling a freemium cloud data storage business

27

Free Premium

π‘₯𝐹 π‘₯𝑃Event β€œCausal” subpopulation Fraction thereof

Recruit Premium users +1 0

Upgrade Free users -1 +1

Downgrade Premium users +1 -1

Attrition Free users -1 0

Page 28: Linear algebra

Example: Modeling a freemium cloud data storage business

28

Free Premium

π‘₯𝐹 π‘₯𝑃

π‘₯𝐹 (𝑑+βˆ† 𝑑 )=π‘₯𝐹 (𝑑 )+𝜌π‘₯ 𝑃 (𝑑 )βˆ’πœπ‘₯𝐹 (𝑑 )+𝛿π‘₯𝑃 (𝑑 )βˆ’π›Όπ‘₯𝐹 (𝑑 )

Event β€œCausal” subpopulation Fraction thereof

Recruit Premium users +1 0

Upgrade Free users -1 +1

Downgrade Premium users +1 -1

Attrition Free users -1 0

π‘₯𝑃 (𝑑+βˆ† 𝑑 )=π‘₯𝑃 (𝑑 )+𝜌 π‘₯𝑃 (𝑑 )+𝜐 π‘₯𝐹 (𝑑 )βˆ’π›Ώπ‘₯ 𝑃 (𝑑 )βˆ’π›Ό π‘₯𝐹 (𝑑 )

Page 29: Linear algebra

Example: Modeling a freemium cloud data storage business

29

Free Premiumπ‘₯𝐹 π‘₯𝑃π‘₯𝐹 (𝑑+βˆ† 𝑑 )=π‘₯𝐹 (𝑑 )+𝜌π‘₯ 𝑃 (𝑑 )βˆ’πœπ‘₯𝐹 (𝑑 )+𝛿π‘₯𝑃 (𝑑 )βˆ’π›Όπ‘₯𝐹 (𝑑 )π‘₯𝑃 (𝑑+βˆ† 𝑑 )=π‘₯𝑃 (𝑑 )+𝜌 π‘₯𝑃 (𝑑 )+𝜐 π‘₯𝐹 (𝑑 )βˆ’π›Ώπ‘₯ 𝑃 (𝑑 )βˆ’π›Ό π‘₯𝐹 (𝑑 )

[π‘₯𝐹 (𝑑+βˆ† 𝑑 )π‘₯𝑃 (𝑑+βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ][π‘₯𝐹 (𝑑 )π‘₯𝑃 (𝑑 ) ]

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

�⃑� (𝑑 )=π‘₯𝐹 (𝑑 ) �⃑� +π‘₯𝑃 (𝑑 ) �⃑�

Page 30: Linear algebra

0 0.25 0.5 0.75 1 1.25 1.5 1.750

0.25

0.5

0.75

1

1.25

1.5

1.75 οΏ½Μ‚οΏ½ �⃑�=πœ†π‘£

Example: Modeling a freemium cloud data storage business

30

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

π‘₯ 𝑃

𝑁

π‘₯ 𝐹

𝑁

Easy-

lookin

g-one-d

imen

siona

l pro

blem

+

Page 31: Linear algebra

Example: Modeling a freemium cloud data storage business

31

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

οΏ½Μ‚οΏ½ �⃑�=πœ†π‘£οΏ½Μ‚οΏ½ �⃑�=πœ† 𝐼 �⃑�

οΏ½Μ‚οΏ½ οΏ½βƒ‘οΏ½βˆ’ πœ† οΏ½Μ‚οΏ½ �⃑�= 0⃑( οΏ½Μ‚οΏ½βˆ’ πœ†πΌ ) �⃑�= 0⃑

([𝔸1 ,1 𝔸1 , 2

𝔸2 , 1 𝔸2 , 2]βˆ’ πœ† [1 00 1])[𝑣𝐹

𝑣𝑃 ]=[00 ]([π‘Ž 𝑏𝑐 𝑑]βˆ’ πœ†[1 0

0 1 ])[𝑣 𝐹

𝑣 𝑃 ]=[00]

STOP

Check that

is consistent in a matrix representation

[π‘Žβˆ’ πœ† 𝑏𝑐 π‘‘βˆ’ πœ†][𝑣𝐹

𝑣 𝑃 ]=[00](π‘Žβˆ’ πœ† )𝑣 𝐹+𝑏𝑣𝑃=0𝑐 𝑣𝐹+(π‘‘βˆ’πœ† )𝑣 𝑃=0

Page 32: Linear algebra

Example: Modeling a freemium cloud data storage business

32

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

οΏ½Μ‚οΏ½ �⃑�=πœ†π‘£(π‘‘βˆ’ πœ† ) (π‘Žβˆ’ πœ† )𝑣𝐹+ (π‘‘βˆ’ πœ† )𝑏𝑣𝑃=0

𝑏𝑐 𝑣𝐹+𝑏 (π‘‘βˆ’ πœ† )𝑣 𝑃=0- [ ][ (π‘Žβˆ’πœ† ) (π‘‘βˆ’ πœ† )βˆ’π‘π‘ ]𝑣𝐹=0

(π‘Žβˆ’ πœ† ) (π‘‘βˆ’ πœ† )βˆ’π‘π‘=0π‘Žπ‘‘βˆ’ πœ†π‘Žβˆ’ πœ†π‘‘+πœ†2βˆ’π‘π‘=0

πœ†2βˆ’ (π‘Ž+𝑑 ) πœ†+(π‘Žπ‘‘βˆ’π‘π‘ )=0

πœ†Β±=(π‘Ž+𝑑 )±√ (π‘Ž+𝑑 )2βˆ’4 (1 ) (π‘Žπ‘‘βˆ’π‘π‘ )

2 (1 )

πœ†Β±=(π‘Ž+𝑑 )Β±βˆšπ‘Ž2+2π‘Žπ‘‘+𝑑2βˆ’4 π‘Žπ‘‘+4𝑏𝑐

2

Page 33: Linear algebra

Example: Modeling a freemium cloud data storage business

33

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

οΏ½Μ‚οΏ½ �⃑�=πœ†π‘£πœ†Β±=

(π‘Ž+𝑑 )Β±βˆšπ‘Ž2+2π‘Žπ‘‘+𝑑2βˆ’4 π‘Žπ‘‘+4𝑏𝑐2

πœ†Β±=(π‘Ž+𝑑 )±√ (π‘Žβˆ’π‘‘ )2+4𝑏𝑐

2

πœ†Β±=(2βˆ’πœβˆ’π›Όβˆ’π›Ώ )±√ (π›Ώβˆ’πœβˆ’π›Ό )2+4 (𝜌+𝛿 )𝜐

2There are 2 possibly special scaling factors. Does each l actually correspond to a special ?

Page 34: Linear algebra

Example: Modeling a freemium cloud data storage business

34

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

οΏ½Μ‚οΏ½ �⃑�=πœ†π‘£πœ†Β±=

(π‘Ž+𝑑 )±√ (π‘Žβˆ’π‘‘ )2+4𝑏𝑐2

There are 2 possibly special scaling factors. Does each l actually correspond to a special ?

[π‘Žβˆ’ πœ†Β± 𝑏𝑐 π‘‘βˆ’ πœ†Β±] [𝑣 𝐹

Β±

𝑣 𝑃± ]=[00]

(π‘Žβˆ’ πœ†Β±) 𝑣𝐹± +𝑏𝑣𝑃

Β±=0𝑏𝑣 𝑃

Β±= (πœ†Β±βˆ’π‘Ž )𝑣𝐹±

𝑣 𝑃±=

πœ†Β±βˆ’π‘Žπ‘ 𝑣 𝐹

Β±

𝑣 𝑃±=

𝛼+πœβˆ’ π›ΏΒ±βˆš (π›Ώβˆ’πœβˆ’π›Ό )2+4 (𝜌+𝛿 )𝜐2 (𝜌+𝛿 )

𝑣𝐹±

Page 35: Linear algebra

Example: Modeling a freemium cloud data storage business

35

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

οΏ½Μ‚οΏ½ �⃑�=πœ†π‘£πœ†Β±=

(π‘Ž+𝑑 )±√ (π‘Žβˆ’π‘‘ )2+4𝑏𝑐2

𝑣 𝑃±=

𝛼+πœβˆ’ π›ΏΒ±βˆš (π›Ώβˆ’πœβˆ’π›Ό )2+4 (𝜌+𝛿 )𝜐2 (𝜌+𝛿 )

𝑣𝐹±

𝑄±

There are 2 special scaling factors; each l corresponds to a special vector . Unless something is hokey, they point in different directions and can serve as a basis.

�⃑�+ΒΏ ΒΏ π‘£βˆ’

Page 36: Linear algebra

Example: Modeling a freemium cloud data storage business

36

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

οΏ½Μ‚οΏ½ 𝑣±=πœ†Β± 𝑣±𝑣 𝑃

Β±=𝑄±𝑣𝐹±

�⃑� (𝑑 )=π‘₯𝐹 (𝑑 ) �⃑� +π‘₯𝑃 (𝑑 ) �⃑�𝑁 �⃑� +0 �⃑�=𝑁𝑐

+ ¿⃑𝑣+ΒΏ+π‘π‘βˆ’π‘£βˆ’ΒΏ ΒΏ

�⃑�+ΒΏ ΒΏ π‘£βˆ’

�⃑�=𝑐+ΒΏ ΒΏΒΏ

Inaugural trials

Page 37: Linear algebra

Example: Modeling a freemium cloud data storage business

37

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

�⃑�=𝑐+ΒΏ ΒΏΒΏ

�⃑�=¿𝑐+ΒΏ+π‘βˆ’=1ΒΏ 𝑐

+¿𝑄 +ΒΏ+𝑐 βˆ’π‘„βˆ’=0ΒΏ ΒΏ

οΏ½Μ‚οΏ½ 𝑣±=πœ†Β± 𝑣±𝑣 𝑃

Β±=𝑄±𝑣𝐹±

�⃑� (𝑑 )=π‘₯𝐹 (𝑑 ) �⃑� +π‘₯𝑃 (𝑑 ) �⃑�𝑁 �⃑� +0 �⃑�=𝑁𝑐

+ ¿⃑𝑣+ΒΏ+π‘π‘βˆ’π‘£βˆ’ΒΏ ΒΏ

Inaugural trials

�⃑�=𝑐+ΒΏ ΒΏΒΏ

Page 38: Linear algebra

Example: Modeling a freemium cloud data storage business

38

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

𝑐+ΒΏ+π‘βˆ’=1ΒΏ 𝑐+¿𝑄 +ΒΏ+𝑐 βˆ’π‘„βˆ’=0ΒΏ ΒΏ

𝑐+¿𝑄 +ΒΏ=βˆ’ π‘βˆ’π‘„βˆ’ΒΏ ΒΏ

𝑐+ΒΏ=βˆ’π‘βˆ’

π‘„βˆ’

𝑄+ΒΏΒΏΒΏβˆ’π‘βˆ’

π‘„βˆ’

𝑄+ΒΏ+π‘βˆ’=1ΒΏ

π‘βˆ’ΒΏπ‘βˆ’ΒΏ π‘βˆ’=

𝑄+ΒΏ

𝑄+ΒΏβˆ’π‘„βˆ’

ΒΏΒΏ 𝑐

+ΒΏ=βˆ’ π‘„βˆ’

𝑄+ΒΏβˆ’π‘„βˆ’

ΒΏΒΏ

�⃑� (𝑑 )=𝑁𝑐+¿⃑ 𝑣+ΒΏ +𝑁 π‘βˆ’π‘£βˆ’ΒΏ ΒΏ

οΏ½Μ‚οΏ½ 𝑣±=πœ†Β± 𝑣±𝑣 𝑃

Β±=𝑄±𝑣𝐹±

Page 39: Linear algebra

Example: Modeling a freemium cloud data storage business

39

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

π‘βˆ’=𝑄+ΒΏ

𝑄+ΒΏβˆ’π‘„βˆ’

ΒΏΒΏ 𝑐

+ΒΏ=βˆ’ π‘„βˆ’

𝑄+ΒΏβˆ’π‘„βˆ’

ΒΏΒΏ

�⃑� (𝑑 )=𝑁𝑐+¿⃑ 𝑣+ΒΏ +𝑁 π‘βˆ’π‘£βˆ’ΒΏ ΒΏ

οΏ½Μ‚οΏ½ 𝑣±=πœ†Β± 𝑣±𝑣 𝑃

Β±=𝑄±𝑣𝐹±

οΏ½Μ‚οΏ½ οΏ½Μ‚οΏ½ οΏ½Μ‚οΏ½ �⃑� (𝑑 )=βˆ’ π‘π‘„βˆ’

𝑄+ΒΏβˆ’π‘„βˆ’

πœ†+ΒΏπœ†+ΒΏ πœ†+ ΒΏ �̂�⃑

𝑣 +ΒΏ+ 𝑁𝑄+ΒΏ

𝑄+ ΒΏβˆ’π‘„βˆ’ πœ†βˆ’ πœ†βˆ’ πœ†βˆ’ οΏ½Μ‚οΏ½ οΏ½βƒ‘οΏ½βˆ’ΒΏΒΏ ΒΏ ΒΏ

ΒΏ ΒΏΒΏ

�̂�𝑀 �⃑� (𝑑 )=βˆ’ π‘π‘„βˆ’

𝑄+ΒΏβˆ’π‘„βˆ’

πœ†+¿𝑀 𝑣+ΒΏ + 𝑁𝑄+ΒΏ

𝑄+ΒΏβˆ’π‘„βˆ’

πœ†βˆ’π‘€π‘£βˆ’ΒΏ

ΒΏΒΏ ΒΏΒΏ

Page 40: Linear algebra

Example: Modeling a freemium cloud data storage business

40

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

οΏ½Μ‚οΏ½ 𝑣±=πœ†Β± 𝑣±𝑣 𝑃

Β±=𝑄±𝑣𝐹±

�̂�𝑀 �⃑� (𝑑 )=βˆ’ π‘π‘„βˆ’

𝑄+ΒΏβˆ’π‘„βˆ’

πœ†+¿𝑀 𝑣+ΒΏ + 𝑁𝑄+ΒΏ

𝑄+ΒΏβˆ’π‘„βˆ’

πœ†βˆ’π‘€π‘£βˆ’ΒΏ

ΒΏΒΏ ΒΏΒΏ

�⃑� (𝑑+π‘€βˆ† 𝑑 )=βˆ’ π‘π‘„βˆ’

𝑄+ΒΏβˆ’π‘„βˆ’

πœ†+¿𝑀¿ ΒΏΒΏ

π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )=𝑁𝑄+ΒΏπœ†βˆ’

π‘€βˆ’π‘„βˆ’ πœ†+ΒΏ 𝑀

𝑄 +ΒΏ βˆ’π‘„βˆ’

ΒΏΒΏ

ΒΏ

π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )=𝑁 𝑄+ΒΏπ‘„βˆ’

𝑄+ΒΏβˆ’π‘„βˆ’

ΒΏΒΏΒΏ

Page 41: Linear algebra

Example: Modeling a freemium cloud data storage business

41

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

οΏ½Μ‚οΏ½ 𝑣±=πœ†Β± 𝑣±𝑣 𝑃

Β±=𝑄±𝑣𝐹±

π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )=𝑁𝑄+ΒΏπœ†βˆ’

π‘€βˆ’π‘„βˆ’ πœ†+ΒΏ 𝑀

𝑄 +ΒΏ βˆ’π‘„βˆ’

ΒΏΒΏ

ΒΏ

π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )=𝑁 𝑄+ΒΏπ‘„βˆ’

𝑄+ΒΏβˆ’π‘„βˆ’

ΒΏΒΏΒΏ

πœ†Β±=(2βˆ’πœβˆ’π›Όβˆ’π›Ώ )±√ (π›Ώβˆ’πœβˆ’π›Ό )2+4 (𝜌+𝛿 )𝜐

2

𝑄±=𝛼+πœβˆ’π›ΏΒ±βˆš (π›Ώβˆ’πœβˆ’π›Ό )2+4 (𝜌+𝛿 )𝜐

2 (𝜌+𝛿 )

, , , = 0.2, 0.2, 0.1, 0.1

Page 42: Linear algebra

0 0.25 0.5 0.75 1 1.25 1.5 1.750

0.25

0.5

0.75

1

1.25

1.5

1.75

Example: Modeling a freemium cloud data storage business

42

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

π‘₯ 𝑃

𝑁

π‘₯ 𝐹

𝑁

οΏ½Μ‚οΏ½ 𝑣±=πœ†Β± 𝑣±

, , , = 0.2, 0.2, 0.1, 0.1

Page 43: Linear algebra

Example: Modeling a freemium cloud data storage business

43

Free

Premium

π‘₯𝐹

π‘₯𝑃

[π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )]=[1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ][1βˆ’πœβˆ’π›Ό 𝜌+π›Ώπœ 1βˆ’π›Ώ]β‹― [1βˆ’πœβˆ’π›Ό 𝜌+𝛿

𝜐 1βˆ’π›Ώ ] [π‘₯ 𝐹 (𝑑 )π‘₯ 𝑃 (𝑑 )]

M copies of matrix

οΏ½Μ‚οΏ½ 𝑣±=πœ†Β± 𝑣±𝑣 𝑃

Β±=𝑄±𝑣𝐹±

π‘₯𝐹 (𝑑+π‘€βˆ† 𝑑 )=𝑁𝑄+ΒΏπœ†βˆ’

π‘€βˆ’π‘„βˆ’ πœ†+ΒΏ 𝑀

𝑄 +ΒΏ βˆ’π‘„βˆ’

ΒΏΒΏ

ΒΏ

π‘₯𝑃 (𝑑+𝑀 βˆ† 𝑑 )=𝑁 𝑄+ΒΏπ‘„βˆ’

𝑄+ΒΏβˆ’π‘„βˆ’

ΒΏΒΏΒΏ

πœ†Β±=(2βˆ’πœβˆ’π›Όβˆ’π›Ώ )±√ (π›Ώβˆ’πœβˆ’π›Ό )2+4 (𝜌+𝛿 )𝜐

2

𝑄±=𝛼+πœβˆ’π›ΏΒ±βˆš (π›Ώβˆ’πœβˆ’π›Ό )2+4 (𝜌+𝛿 )𝜐

2 (𝜌+𝛿 )

Eigenvectors

Eigenvalues

, , , = 0.2, 0.2, 0.1, 0.1