Optimization and

algorithms, IST &

RC 2022, Joo Xavier

Solution of Exam1-Part A

Problem 1 Answer:A

ails
#

&
Note that s B(c) if and only if Y&B(c). That is,
the ball B(c) does

not covers ifand only if there exists a victor vel much that
uckB(c).
Thus: -
fB() ( 7:vB(c)
vev

A 7: llv-cll2 > r

rev

v=vc
#Fitz: r

Option A

&
Option B is incorrect because it allows

⑤ ⑤

⑤ ⑤

&
Option C is incorrect because it forbids

⑤ ⑧

*
Option D is incorrect because it allows

· &

⑳ ⑳
 &
Option E is incorrect because it allows the configuration shown in

* F incorrect because configuration B

Option is it allows the shown in

Problem 2 Answer: E

fails
&

This problem is
convix, and this function

f(x) (ax b)-

= -
+ ((x1lz c)-
+ 1x- dk

is differentiable at x4 =
[5].

&
So, xt is a
global minimizer for fit and only if

-f(x+) = 0(-) a(ax+ -

b)a +
2(lx+1lz c) + -

+ 2(x+ -

d) = 0

-d
xE = (aTx+ -
b)a +
(lx+1lz -
c)
+ le + x*

(tiy[] x() Hz
) 1E) E)
c
v = - +
-

+
+
wo
O

a v
E) [3]
= +

a d
[
=
 Problicm 3 Answer: B

tails
#

call that of the form

an is problem is an optimization problism
min.
X
11Ax-BIEc
for some matrix A and victor
p

In option
& A, the map raw: 1R"-i" or a linor map, that is, it can

be written in the form rev(x) =

Mx, for some matrix M.. Indeed, taking

have
he
->
special Cask nit, we

wr(()) I re
=
O 0

-
X

This means that

(Arev(x)-bIE T
HIEMx-b that's
a
mn.
min. ↑

25
problem
This
& above also to options C, D, E, and F; only
reasoning applies the matrix

will
change:

circlino.
O 00

·orine

·
orED
cant.k-KEEABEYEStrit7 EYEEEEE
So FENNOErio
I I

I
O I
=
1 O

B O

I fails
*
. .

&
In the Newton algorithm xkx=xx+CkOn, the direction di in given by

dk = - x2f(x)" +f(x)

For
& F(a,b) = (2a-b1"+b", we have

wfsa,b) =

[8a-0bv -+cab)
13)
=

Evaluating
& of and UF at xx:
1) gives

of(x)
[i]
= cr (xa)=
F
Thus,
* du =
-

E"[i] Ez =
& ptimization and
algorithms, IST &
icc 2022, Joao Xavier

Solution of Exam 1 -
Part B

Problem

The problem
& min.
X ox-bx) + xP(lx -

alle r)
-

f(x)

S.t. xx(ly - r = O

g(x)

is convex if fand are convex, which we now show

* this of (a well-known function)

NX-G is rum a norm convex

with a constant (which does not affect convexity

· So, I is convex

convex
& fas nonmigative linear combination of F-,fre, andh:
write
o

f f =
+ Fx+-.+ Fx +
xh,

where f(x) (anx b1) and h(x) N(((x r)

= =

cl(c
-
- -

· if we show
fra, fr, h are
convex, the function of will be convix

·
Esconvox this the
composition to PR

where paax-anix-bae is affine and

is
& convex (it suffices to look at

dis
of
tungropk
 ·convex his the composition Noq, where

↑
is convex and non-decreasing (as obvious

i
from its groom

anda is convox:
9 is the
composition
o a convex function, 11-12, with the affin

map
X to x-c (the additive constant -
r

is irrelevant for the

convexity of al

Problem 2 Alice is
right:
R

Bob's problem
& is
man. Slx-P1*
-
fBoB(x)

Mt. sTx
V,
=

S2X-E)
=x

S
~ hose RRT conditions are wf3(x) = 3x
+

Six r
=

where F = Ph R
2
1)x p1l2
⑫ Alick's problem is min. -

X an

fALicE(x)
sit. Six rc
=

L j
whose RRT conditions are -> fALice(x) = 5 ,
-
2(x -

p) sx
=

gx r =

sx = r

&
The MRT conditions of Bob and Alice are the
same; so, the minimizers are the same
 H= Ex: sTx r,y
=

Problem 3
Hel H2 Gx= sTx vet
**
=
=

Yoch,
The goal
* is to find dLH., He): ⑤

The
& quantity (CHi, Hz) is equal to 11x.*-xetle whenever x.*, xe* are

solutions of the problem

min- 1y((x -
xe11?
*
1, X2 -
f(x),xz)

sof six -

r
0
=

n,(x,,xe)

87x2 -
r 0
=

he(X1,Xz)

whose RRT conditions are

S S
(i =[ []i
(x)(i)

4
The(xyxe)X2
-E
-f(x,xz) Th(x1Xz)x, + X. x2 =
-
=

xz -
x1 =
5x2 (ii)
h(xx(z) 0
= >
->

six. The 53x, Y

=

(iii)
hz(X,,xe) 0
=

six2 Vz =

(iv)

&
Subtract (iii) ono) (iv) to get
x(x, -
xz) V
=
-

rz,

(from (il) to at
and
plug X.-X2TSX. arrive

ssx, v rz- x=
rin
-
=

*
using itin cil gives x1-42:
the
are

*
This means (CHi, He) =
11x1 -xell =

re
Problem 4

The
& problem is min. (lX -

cz -
B)
X
-
f(x)

5.t. six -
r 0
=

h(x]

whosi RRT conditions are

cz
b
Bb +

S
- f(x) yh(x)x
= 2(1x -
c -
+ sT
-

h(x) 0
=
six r
=

Two
& cases can arisi:
(((x-cl1-B)+=0 or (1x-(11-B)+>0

③ - c(1 -

B) + =0 this
·
means that we are
invistigating the existinia

of a minimizerx that satisfies lx-cle-B,

that
is, x is in the ball B: Eu: 1ln-c(12 -
B

· In this case, KKC system becomes

T
s#0)
(- x =0 becouri

3
· is x

six v,(E
=
-
xc H)

*
with solutions being all pairs (x,x) retisfying

AB and x= 0

=
*
case (((x c((z B) +30
- -
· In this CaSI, RKT system becomes
- -

cTc
S
2(lx cz B3 - - = sT (i)

six r
= (ii)
 ·
(i) implies

x-

c3
i s,
y
that is, x is of the form x=c+ys, for someyCIR

into (ii) to
get
plug +ys
·
x- c

stx + ys) =
rz
y
=

5s
c

& "B which means

such
x = c +
5trans
*
a reborn is a solution in ause it also

satisfies x*B