Transcript
Page 1: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

On Convexity of Polynomials over a Box

Georgina HallDecision Sciences, INSEAD

Joint work withAmir Ali Ahmadi

ORFE, Princeton University

1

Page 2: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Convexity over a boxโ€ข A box ๐‘ฉ is a set of the form:

๐ต = ๐‘ฅ โˆˆ โ„๐‘› ๐‘™๐‘– โ‰ค ๐‘ฅ๐‘– โ‰ค ๐‘ข๐‘– , ๐‘– = 1, โ€ฆ , ๐‘›}

where ๐‘™1, โ€ฆ , ๐‘™๐‘›, ๐‘ข1, โ€ฆ , ๐‘ข๐‘› โˆˆ โ„ with ๐‘™๐‘– โ‰ค ๐‘ข๐‘– .

โ€ข A function ๐’‡ is convex over ๐‘ฉ if ๐‘“ ๐œ†๐‘ฅ + 1 โˆ’ ๐œ† ๐‘ฆ โ‰ค ๐œ†๐‘“ ๐‘ฅ + 1 โˆ’ ๐œ† ๐‘“(๐‘ฆ)

for any ๐‘ฅ, ๐‘ฆ โˆˆ ๐ต and ๐œ† โˆˆ [0,1].

โ€ข If ๐‘ฉ is full dimensional (i.e., ๐‘™๐‘– < ๐‘ข๐‘–, ๐‘– = 1,โ€ฆ , ๐‘›), this is equivalent to

๐›ป2๐‘“ ๐‘ฅ โ‰ฝ 0, โˆ€๐‘ฅ โˆˆ ๐ต.

2

Page 3: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Complexity questions

โ€ข Restrict ourselves to polynomial functions.

โ€ข Related work:

Theorem [Ahmadi, Olshevsky, Parrilo, Tsitsiklis]It is strongly NP-hard to test (global) convexity of polynomials of degree 4.

โ€ข One may hope that adding the restriction to a box could make things easier.

3

Goal: study the complexity of testing convexity of a function over a box

Page 4: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Our theorem

Why are we interested in convexity over a box?

4

Theorem [Ahmadi, H.]It is strongly NP-hard to test convexity of polynomials of degree 3 over a box.

โ€ข Nonconvex optimization: branch-and-bound

โ€ข Prior work: โ€ข Sufficient conditions for convexity [Orban et

al.], [Grant et al.]โ€ข In practice, BARON, CVX, Gurobi check

convexity of quadratics and computationally tractable sufficient conditions for convexity

Detecting Imposing

โ€ข Control theory: convex Lyapunov functions

โ€ข Statistics: convex regression

[Ahmadi and Jungers][Chesi and Hung]

Page 5: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Question: What to do a reduction

from?

Idea: A cubic polynomial ๐‘“ is convexover a (full-dimensional) box ๐ต if and

only if ๐›ป2๐‘“ ๐‘ฅ โ‰ฝ 0, โˆ€๐‘ฅ โˆˆ ๐ต

๐›ป2๐‘“(๐‘ฅ) is a matrix with entries affine

in ๐’™

Proof of the theorem

5

Theorem [Ahmadi, H.]It is strongly NP-hard to test convexity of polynomials of degree 3 over a box.

How to prove this?

In general:

Theorem [Nemirovski]:Let ๐ฟ(๐‘ฅ) be a matrix with entries affine in ๐‘ฅ.

It is NP-hard to test whether ๐ฟ ๐‘ฅ โ‰ฝ 0 for all ๐‘ฅ in a full-dimensional box ๐ต.

Generic instance I of a known

NP-hard problem

Instance J of problem we are

interested in

Construct J from I

Reduction

Page 6: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Are we done?No!

6

Issue 1: We want to show strong NP-hardness. Nemirovskiโ€™s result shows weak NP-hardness.

Issue 2: Not every affine polynomial matrix is a valid Hessian!

Example: ๐ฟ ๐‘ฅ1, ๐‘ฅ2 =10 2๐‘ฅ1 + 1

2๐‘ฅ1 + 1 10. We have

๐œ•๐ฟ11(๐‘ฅ)

๐œ•๐‘ฅ2โ‰ 

๐œ•๐ฟ12(๐‘ฅ)

๐œ•๐‘ฅ1.

Page 7: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Dealing with Issue 1 (1/5)Reminder: weak vs strong NP-hardness

โ€ข Distinction only concerns problems where input is numerical

โ€ข Max(I): largest number in magnitude that appears in the input of instance I (numerator or denominator)

โ€ข Length(I): number of bits it takes to write down input of instance I

7

Strong Weak

โ€ข There are instances ๐ผ that are hard with Max(๐ผ)โ‰ค ๐‘(Length(๐ผ)) (๐‘ is a polynomial)

โ€ข No pseudo-polynomial algorithm possible

โ€ข Examples:

Max-Cut

Sat

โ€ข The instances that are hard may contain numbers of large magnitude (e.g., 2๐‘›).

โ€ข Pseudo-polynomial algorithms possible

โ€ข Examples:

Partition

Knapsack

Page 8: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Dealing with Issue 1 (2/5)

Why weakly NP-hard?

8

Theorem [Nemirovski]: INTERVAL-PSDNESSLet ๐ฟ(๐‘ฅ) be a matrix with entries affine in ๐‘ฅ.

It is (weakly) NP-hard to test whether ๐ฟ ๐‘ฅ โ‰ฝ 0 for all ๐‘ฅ in a full-dimensional box ๐ต.

PARTITION:

Input: ๐‘Ž โˆˆ โ„๐‘› such that ๐‘Ž 2 โ‰ค 0.1

Test: does there exist ๐‘ก โˆˆ โˆ’1,1 ๐‘›

such that ฯƒ๐‘– ๐‘Ž๐‘–๐‘ก๐‘– = 0?

INTERVAL PSDNESS

Construct: ๐ถ = ๐ผ๐‘› โˆ’ ๐‘Ž๐‘Ž๐‘‡โˆ’1,

๐œ‡ = ๐‘› โˆ’ ๐‘‘โˆ’2 ๐‘Ž , where ๐‘‘ ๐‘Ž = smallest cd of ๐‘Ž.

Take: ๐ต = โˆ’1,1 ๐‘› and ๐ฟ ๐‘ฅ =๐ถ ๐‘ฅ๐‘ฅ๐‘‡ ๐œ‡

.

Test: Is ๐ฟ ๐‘ฅ โ‰ฝ 0 โˆ€๐‘ฅ โˆˆ ๐ต?Show: No to PARTITION โ‡” Yes to INTERVAL PSDNESS

REDUCTION

Weakly NP-hardOperation that can make the numbers in the instance blow up

Example: ๐ด =

1 0 0 0โˆ’1 1 0 0โ‹ฎโˆ’1

โ‹ฑโˆ’1

โ‹ฑโˆ’1

01

but one of the entries of ๐ดโˆ’1 is 2๐‘›โˆ’2!=๐‘Ž1๐‘Ž3๐‘Ž10 ๐‘Ž2 ๐‘Ž4๐‘Ž8

Page 9: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Dealing with Issue 1 (3/5)

9

Theorem [Ahmadi, H.]: INTERVAL-PSDNESSLet ๐ฟ(๐‘ฅ) be a matrix with entries affine in ๐‘ฅ.

It is strongly NP-hard to test whether ๐ฟ ๐‘ฅ โ‰ฝ 0 for all ๐‘ฅ in a full-dimensional box ๐ต.

MAX-CUT:

Input: simple graph G=(V,E) with ๐‘‰ = ๐‘› and adj. matrix A, and a

positive integer ๐‘˜ โ‰ค ๐‘›2

Test: does there exist a cut in the graph of size greater or equal to ๐‘˜?

INTERVAL PSDNESS

Construct: ๐›ผ =1

๐‘›+1 3 , ๐ถ = 4๐›ผ(๐ผ๐‘› + ๐›ผ๐ด)

๐œ‡ =๐‘›

4๐›ผ+ ๐‘˜ โˆ’ 1 โˆ’

1

4๐‘’๐‘‡๐ด๐‘’

Take: ๐ต = โˆ’1,1 ๐‘› and ๐ฟ ๐‘ฅ =๐ถ ๐‘ฅ๐‘ฅ๐‘‡ ๐œ‡

.

Test: Is ๐ฟ ๐‘ฅ โ‰ฝ 0 โˆ€๐‘ฅ โˆˆ ๐ต?Show: No to MAX-CUT โ‡” Yes to INTERVAL PSDNESS

REDUCTION

Strongly NP-hard Taylor series of 4๐›ผ ๐ผ โˆ’ ๐›ผ๐ด โˆ’1 truncated at the first term

Scaling needed so that ๐ผ๐‘› โˆ’ ๐›ผ๐ด โˆ’1 โ‰ˆ ๐ผ๐‘› + ๐›ผ๐ด

Preserves strong NP-hardness

Page 10: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Dealing with Issue 1 (4/5)In more detail: No to MAX-CUT โ‡’ Yes to INTERVAL PSDNESS

10

No cut in ๐บ of size โ‰ฅ ๐‘˜ [max๐‘ฅโˆˆ โˆ’1,1 ๐‘›

1

4ฯƒ๐‘–,๐‘— ๐ด๐‘–๐‘—(1 โˆ’ ๐‘ฅ๐‘–๐‘ฅ๐‘—)] โ‰ค ๐‘˜ โˆ’ 1

Size of largest cut in ๐บ

โ‡”

โ‡”

[ max๐‘ฅโˆˆ โˆ’1,1 ๐‘›

โˆ’1

4๐‘ฅ๐‘‡๐ด๐‘ฅ] โ‰ค โˆ’

1

4๐‘’๐‘‡๐ด๐‘’ + ๐‘˜ โˆ’ 1โ‡”[ max

๐‘ฅโˆˆ โˆ’1,1 ๐‘›

1

4๐‘ฅ๐‘‡ ๐‘› + 1 3๐ผ๐‘› โˆ’ ๐ด ๐‘ฅ] โ‰ค

๐‘› ๐‘›+1 3

4โˆ’

1

4๐‘’๐‘‡๐ด๐‘’ + ๐‘˜ โˆ’ 1 โ‰” ๐œ‡

๐›ผ = ๐‘› + 1 3

Convex

โ‡”

[ max๐‘ฅโˆˆ[โˆ’1,1]๐‘›

1

4๐‘ฅ๐‘‡ ๐›ผ๐ผ๐‘› โˆ’ ๐ด ๐‘ฅ] โ‰ค ๐œ‡ โ‡”

1

4๐‘ฅ๐‘‡ ๐›ผ๐ผ๐‘› โˆ’ ๐ด ๐‘ฅ โ‰ค ๐œ‡, โˆ€๐‘ฅ โˆˆ โˆ’1,1 ๐‘›

โ‡’

๐‘ฅ๐‘‡๐ถโˆ’1๐‘ฅ โ‰ค ๐œ‡ +1

4, โˆ€๐‘ฅ โˆˆ โˆ’1,1 ๐‘›

Approximation ๐ถโˆ’1 โ‰ˆ1

4(๐›ผ๐ผ โˆ’ ๐ด)

Approximation error

โ‡’

Schurcomplement

๐ฟ ๐‘ฅ =๐ถ ๐‘ฅ

๐‘ฅ๐‘‡ ๐œ‡ +1

4

โ‰ฝ 0, โˆ€๐‘ฅ โˆˆ โˆ’1,1 ๐‘›

Page 11: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Dealing with Issue 1 (5/5)For converse: Yes to MAX-CUT โ‡’ No to INTERVAL PSDNESS

โ€ข Initial problem studied by Nemirovski

โ€ข Of independent interest in robust control

11

There is a cut of size โ‰ฅ ๐‘˜:

Let เทœ๐‘ฅ๐‘– = แ‰Š1 if node ๐‘– on one side of cut

โˆ’1 if node ๐‘– on other side of cutโ‡’ Similar steps

to previously โ‡’ เทœ๐‘ฅ๐‘‡๐ถโˆ’1 เทœ๐‘ฅ โ‰ฅ ๐œ‡ +3

4> ๐œ‡ +

1

4

โˆƒ เทœ๐‘ฅ โˆˆ โˆ’1,1 ๐‘› s.t. ๐ฟ เทœ๐‘ฅ 0โ‡’

Corollary [Ahmadi, H.]: Let ๐‘› be an integer and let เทœ๐‘ž๐‘–๐‘— , เดค๐‘ž๐‘–๐‘— be rational numbers

with เทœ๐‘ž๐‘–๐‘— โ‰ค เดค๐‘ž๐‘–๐‘— and เทœ๐‘ž๐‘–๐‘— = เทœ๐‘ž๐‘—๐‘– and เดค๐‘ž๐‘–๐‘— = เดค๐‘ž๐‘—๐‘– for all ๐‘– = 1,โ€ฆ , ๐‘› and ๐‘— = 1,โ€ฆ , ๐‘›.

It is strongly NP-hard to test whetherall symmetric matrices with entries in [ เทœ๐‘ž๐‘–๐‘—; เดค๐‘ž๐‘–๐‘—] are positive semidefinite.

Page 12: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Dealing with Issue 2 (1/3)

Proof: Reduction from INTERVAL PSDNESS

12

Theorem [Ahmadi, H.] CONV3BOXIt is strongly NP-hard to test convexity of polynomials of degree 3 over a box.

INTERVAL PSDNESSInput: ๐ฟ ๐‘ฅ , ๐ต

Test: Is ๐ฟ ๐‘ฅ โ‰ฝ 0, โˆ€๐‘ฅ โˆˆ ๐ต?

Problem: How to construct a cubic polynomial ๐‘“ from ๐ฟ(๐‘ฅ)?Idea: Want ๐›ป2๐‘“ ๐‘ฅ = ๐ฟ ๐‘ฅ .Issue: Not all ๐ฟ(๐‘ฅ) are valid Hessians!

Key ideas for the construction of ๐’‡:

โ€ข Start with ๐’‡ ๐’™, ๐’š =๐Ÿ

๐Ÿ๐’š๐‘ป๐‘ณ ๐’™ ๐’š

โ€ข For ๐›ป2๐‘“ ๐‘ฅ, ๐‘ฆ to be able to be psd when ๐ฟ ๐‘ฅ โ‰ฝ 0 , we need to have

a nonzero diagonal: add ๐œถ

๐Ÿ๐’™๐‘ป๐’™ to ๐‘“ ๐‘ฅ, ๐‘ฆ .

โ€ข ๐ฟ ๐‘ฅ and ๐ป(๐‘ฆ) do not depend on the same variable: what if โˆƒ(๐‘ฅ, ๐‘ฆ) s.t. ๐ฟ ๐‘ฅ = 0 but ๐ป ๐‘ฆ is not? The matrix cannot be psd: add ๐œ‚

2๐‘ฆ๐‘‡๐‘ฆ to ๐‘“ ๐‘ฅ, ๐‘ฆ .

๐›ป2๐‘“ ๐‘ฅ, ๐‘ฆ =0

1

2๐ป(๐‘ฆ)

1

2๐ป ๐‘ฆ ๐‘‡ ๐ฟ ๐‘ฅ

๐›ป2๐‘“ ๐‘ฅ, ๐‘ฆ =๐œถ๐‘ฐ๐’

1

2๐ป(๐‘ฆ)

1

2๐ป ๐‘ฆ ๐‘‡ ๐ฟ ๐‘ฅ

๐›ป2๐‘“ ๐‘ฅ, ๐‘ฆ =๐œถ๐‘ฐ๐’

1

2๐ป(๐‘ฆ)

1

2๐ป ๐‘ฆ ๐‘‡ ๐ฟ ๐‘ฅ + ๐œ‚๐ผ๐‘›+1

โ‡’ ๐‘“ ๐‘ฅ =1

2๐‘ฆ๐‘‡๐ฟ ๐‘ฅ ๐‘ฆ +

๐›ผ

2๐‘ฅ๐‘‡๐‘ฅ +

๐œ‚

2๐‘ฆ๐‘‡๐‘ฆ, ๐ต = โˆ’1,1 2๐‘›+1

Page 13: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Dealing with Issue 2 (2/3)Show NO to INTERVAL PSDNESS โ‡’ NO to CONV3BOX.

This is equivalent to:

Need to leverage extra structure of ๐ฟ ๐‘ฅ : ๐ฟ ๐‘ฅ =๐ถ ๐‘ฅ

๐‘ฅ๐‘‡ ๐œ‡ +1

4

13

โˆƒ าง๐‘ฅ โˆˆ โˆ’1,1 ๐‘› s.t. ๐ฟ าง๐‘ฅ โ‰ฝ 0 โ‡’ โˆƒ เทœ๐‘ฅ, เทœ๐‘ฆ โˆˆ โˆ’1,1 2๐‘›+1 , ๐‘ง s.t. ๐‘ง๐‘‡๐›ป2๐‘“ เทœ๐‘ฅ, เทœ๐‘ฆ ๐‘ง < 0

๐›ป2๐‘“ ๐‘ฅ, ๐‘ฆ =

๐›ผ๐ผ๐‘› ๐ป(๐‘ฆ)

๐ป ๐‘ฆ ๐‘‡๐ถ + ๐œ‚๐ผ๐‘› ๐‘ฅ

๐‘ฅ๐‘‡ ๐œ‡ +1

4+ ๐œ‚

๐›ป2๐‘“ เทœ๐‘ฅ, เทœ๐‘ฆ =

๐›ผ๐ผ๐‘› ๐ŸŽ ๐ŸŽ๐ŸŽ ๐‘ช + ๐œ‚๐ผ๐‘› เดฅ๐’™

๐ŸŽ เดฅ๐’™๐‘ป ๐ +๐Ÿ

๐Ÿ’+ ๐œ‚

๐‘ง๐‘‡๐›ป2๐‘“ เทœ๐‘ฅ, เทœ๐‘ฆ ๐‘ง =0

โˆ’๐ถโˆ’1 าง๐‘ฅ1

๐‘‡ ๐›ผ๐ผ๐‘› ๐ŸŽ ๐ŸŽ๐ŸŽ ๐‘ช + ๐œ‚๐ผ๐‘› เดฅ๐’™

๐ŸŽ เดฅ๐’™๐‘ป ๐ +๐Ÿ

๐Ÿ’+ ๐œ‚

0โˆ’๐ถโˆ’1 าง๐‘ฅ

1= ๐œ‡ +

1

4โˆ’ าง๐‘ฅ๐‘‡๐ถโˆ’1 าง๐‘ฅ + ๐œ‚(1 + ๐ถโˆ’1 าง๐‘ฅ

2

2)

< ๐ŸŽ as ๐‘ณ เดฅ๐’™ โ‰ฝ ๐ŸŽAppropriately scaled so that ๐‘ง๐‘‡๐›ป2๐‘“ เทœ๐‘ฅ, เทœ๐‘ฆ ๐‘ง remains <0.

Candidates: เทœ๐‘ฅ = าง๐‘ฅ, เทœ๐‘ฆ = 0, ๐‘ง =0

โˆ’๐ถโˆ’1 าง๐‘ฅ1

Candidates: เท๐’™ = เดฅ๐’™, เท๐’š = ๐ŸŽ, ๐‘ง =0

โˆ’๐ถโˆ’1 าง๐‘ฅ1

Candidates: เทœ๐‘ฅ = าง๐‘ฅ, เทœ๐‘ฆ = 0, ๐’› =๐ŸŽ

โˆ’๐‘ชโˆ’๐Ÿเดฅ๐’™๐Ÿ

Page 14: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Dealing with Issue 2 (3/3)Show YES to INTERVAL PSDNESS โ‡’ YES to CONV3BOX.

This is equivalent to:

Butโ€ฆ

14

๐ฟ ๐‘ฅ โ‰ฝ 0 โˆ€๐‘ฅ โˆˆ โˆ’1,1 ๐‘› โ‡’ ๐›ป2๐‘“ ๐‘ฅ, ๐‘ฆ =๐›ผ๐ผ๐‘›

1

2๐ป(๐‘ฆ)

1

2๐ป ๐‘ฆ ๐‘‡ ๐ฟ ๐‘ฅ + ๐œ‚๐ผ๐‘›+1

โ‰ฝ 0, โˆ€ ๐‘ฅ, ๐‘ฆ โˆˆ โˆ’1,1 2๐‘›+1

โ‡”

โ‰ฝ ๐ŸŽโˆ€๐’™ โˆˆ โˆ’๐Ÿ, ๐Ÿ ๐’

(Assumption)

๐‘ณ ๐’™ + ๐œผ๐‘ฐ๐’+๐Ÿ โˆ’๐Ÿ

๐Ÿ’๐œถ๐‘ฏ ๐’š ๐‘ป๐‘ฏ ๐’š โ‰ฝ 0, โˆ€ ๐‘ฅ, ๐‘ฆ โˆˆ โˆ’1,1 2๐‘›+1

๐œถ chosen large enough so thatโ‰ฝ ๐ŸŽ โˆ€๐’š โˆˆ โˆ’๐Ÿ, ๐Ÿ ๐’+๐Ÿ

๐›ป2๐‘“ ๐‘ฅ, ๐‘ฆ โ‰ฝ 0, โˆ€ ๐‘ฅ, ๐‘ฆ โˆˆ โˆ’1,1 2๐‘›+1Schur

Page 15: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

CorollaryCompletely classifies the complexity of testing convexity of a polynomial ๐‘“ of degree ๐‘‘ over a box for any integer ๐‘‘ โ‰ฅ 1.

15

๐’…๐‘‘ = 1

๐‘“ is alwaysconvex

๐‘‘ = 2

๐›ป2๐‘“(๐‘ฅ)constant

๐‘‘ = 3

Previous theorem (strongly NP-hard)

๐‘‘ = 4 and above

Strongly NP-hardProof sketch: โ€ข ๐‘” ๐‘ฅ1, โ€ฆ , ๐‘ฅ๐‘› =cubic polynomial for which testing

convexity over a box ๐ต is hard

โ€ข ๐‘“ ๐‘ฅ1, โ€ฆ , ๐‘ฅ๐‘›, ๐‘ฅ๐‘›+1 = ๐‘” ๐‘ฅ1, โ€ฆ , ๐‘ฅ๐‘› + ๐‘ฅ๐‘›+1๐‘‘

โ€ข เทจ๐ต = ๐ต ร— [0,1]

We have ๐›ป2๐‘“ ๐‘ฅ, ๐‘ฅ๐‘›+1 =๐›ป2๐‘”(๐‘ฅ) 0

0 ๐‘‘ ๐‘‘ โˆ’ 1 ๐‘ฅ๐‘›+1๐‘‘โˆ’2

โ‡’ ๐›ป2๐‘“ ๐‘ฅ, ๐‘ฅ๐‘›+1 โ‰ฝ 0 on เทจ๐ต โ‡” ๐›ป2๐‘” ๐‘ฅ โ‰ฝ 0 on ๐ต

Page 16: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Summaryโ€ข Interested in testing convexity of a polynomial over a box.

โ€ข Showed that strongly NP-hard to test convexity of cubics over a box.

โ€ข Gave a complete characterization of the complexity of testing convexity over a box depending on the degree of the polynomial.

โ€ข In the process, strengthened a result on the complexity of testing positive semidefiniteness of symmetric matrices with entries belonging to intervals.

16

Page 17: On Convexity of Polynomials over a Box...of Polynomials over a Box Georgina Hall Decision Sciences, INSEAD Joint work with Amir Ali Ahmadi ORFE, Princeton University 1 Convexity over

Thank you for listeningQuestions?

Want to learn more?

https://scholar.princeton.edu/ghall

17


Top Related