Report my confusions/errors when finding a feasible solution of a simple LMI given in the tutorials page

Can Li

unread,

Jul 19, 2017, 9:37:47 PM7/19/17

to YALMIP

Hi Johan,

I have some confusions on the simple example of LMI feasible solution given in https://yalmip.github.io/tutorial/semidefiniteprogramming/

when i use the same code (copied from the above link page) posted below:

A = [-1 2 0; -3 -4 1; 0 0 -2];

P = sdpvar(3,3);

F = [P >= 0, A'*P+P*A <= 0];

optimize(F); % for a feasible solution only

Pfeasible = value(P)

It reports errors:

error using: lmi/subsref (line 76)

Indexing type not supported

error in: optimize (line 14)

    diagnostics =

    solvesdp(OptimizationProblem.Constraints,OptimizationProblem.Objective,OptimizationProblem.Options);

error in: sdpex (line 17)

optimize(F);

The same errors happen when i add an objective function as

A = [-1 2 0; -3 -4 1; 0 0 -2];

P = sdpvar(3,3);

F = [P >= 0, A'*P+P*A <= 0];

obj = -P;

optimize(F,obj);

Pfeasible = value(P)

But it works as i try (replace 'optimize' by 'solvesdp')

A = [-1 2 0; -3 -4 1; 0 0 -2];

P = sdpvar(3,3);

F = [P >= 0, A'*P+P*A <= 0];

obj = -P;

solvesdp(F,obj);

Pfeasible = value(P)

or

A = [-1 2 0; -3 -4 1; 0 0 -2];

P = sdpvar(3,3);

F = optproblem([P >= 0, A'*P+P*A <= 0], -P);

optimize(F);

Pfeasible = value(P)

this two codes give the same result

Pfeasible =

    1.0000    0.0723   -0.0091

    0.0723    0.3667    0.0524

   -0.0091    0.0524    0.5725

My problem is: why the first two codes do not work and report errors? Because they are copied from https://yalmip.github.io/tutorial/semidefiniteprogramming/

Should i always add an objective function even if i only need a feasible solution of LMIs?

Thanks for your attention!

Johan Löfberg

unread,

Jul 20, 2017, 3:22:47 AM7/20/17

to YALMIP

You've messed something up when reporting the error or something. The error you show is happening in optproblem/optimize, but you claim the optproblem approach works

My guess is that you have some weird mix of old and new variables or something. Make sure you clean your work-space before running the test

Can Li

unread,

Jul 20, 2017, 4:22:03 AM7/20/17

to YALMIP

Thanks! i try it by adding ‘clear all’ at the beginning of the code. But the errors still exist.

clear all

A = [-1 2 0; -3 -4 1; 0 0 -2];

P = sdpvar(3,3);

F = [P >= 0, A'*P+P*A <= 0];

optimize(F);

Pfeasible = value(P)

error using: lmi/subsref (line 76)

Indexing type not supported

error in: optimize (line 14)

    diagnostics =

    solvesdp(OptimizationProblem.Constraints,OptimizationProblem.Objective,OptimizationProblem.Options);

error in: sdpex (line 5)

optimize(F);

Johan Löfberg

unread,

Jul 20, 2017, 4:48:56 AM7/20/17

to YALMIP

It still goes into an optproblem method (line 14 references line 14 in optproblem/optimize)

You must have something massively wrong with your installation

Run and report

yalmip('ver')

which optimize -all

path

type sdpex.m

Can Li

unread,

Jul 20, 2017, 5:01:23 AM7/20/17

to YALMIP

The information are posted below ('path' follows too many items, so i just give that relate to YALMIP)

>> yalmip('ver')

ans =

20170626

>> which optimize -all

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\extras\optimize.m

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\extras\@optproblem\optimize.m % optproblem method

C:\Program Files\MATLAB\R2016a\toolbox\mbc\mbcview\@cgnormnode\optimize.m % cgnormnode method

C:\Program Files\MATLAB\R2016a\toolbox\sldo\sloptim\sloptim\@ResponseOptimizer\optimize.m % ResponseOptimizer method

>> path

MATLABPATH

D:\文档\MATLAB

D:\文档\MATLAB\sedumi

D:\文档\MATLAB\sedumi\conversion

D:\文档\MATLAB\sedumi\doc

D:\文档\MATLAB\sedumi\examples

D:\文档\MATLAB\sedumi\o_win

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\modules\sos

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\modules\robust

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\modules\global

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\modules\moment

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\modules\parametric

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\operators

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\modules

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\solvers

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\demos

C:\Program Files\MATLAB\R2016a\toolbox\yalmip

C:\Program Files\MATLAB\R2016a\toolbox\yalmip\extras

>> type sdpex.m

clear all

A = [-1 2 0; -3 -4 1; 0 0 -2];

P = sdpvar(3,3);

F = optproblem([P >= 0, A'*P+P*A <= 0], -(P));

% F = [P >= 0, A'*P+P*A <= 0];

optimize(F);

Pfeasible = value(P)

Johan Löfberg

unread,

Jul 20, 2017, 5:25:09 AM7/20/17

to YALMIP

You said if failed when you used the standard approach (I never use optprolem and it is essentially an obsolete/unofficial feature), but in your code you are using optproblem, so you are still not consistently reporting what you are doing and what is happening

Before proceeding, restart matlab. I've never seen this error, and it makes no sense

Change you sdpex to this, run and report everything displayed

clear all
A = [-1 2 0; -3 -4 1; 0 0 -2];
P = sdpvar(3,3);

F = [P >= 0, A'*P+P*A <= 0]
class(F)
methods(F)
optimize(F,[],sdpsettings('debug',1))

...and having -P as objective in your optproblem does not make sense (it should be a scalar)

Can Li

unread,

Jul 20, 2017, 6:20:26 AM7/20/17

to YALMIP

Here is the result displayed: (still report errors)

>> sdpex

++++++++++++++++++++++++++++++++

| ID| Constraint|

++++++++++++++++++++++++++++++++

| #1| Matrix inequality 3x3|

| #2| Matrix inequality 3x3|

++++++++++++++++++++++++++++++++

ans =

lmi

Methods for class lmi:

Polyhedron

and

assignschur

boundingbox

categorizeproblem

chanceconstraint

chebyball

check

checkset

clear_poly_dep

cleardual

colon

complements

constraintclass

convertlorentz

convertsocp

cut

depends

display

dissect

double

dual

eliminateBinary

ellipsoid

end

envelope

eq

expanded

expandmeta

extractRandomDefinitions

fastcat

flatten

getComplementarityTerms

getbase

getbounds

getbounds_interval

getcutflag

getlmiid

getlrdata

getvariables

groupchanceconstraints

horzcat

hull

imag2reallmi

imagemodel

indicators

is

isfeasible

isinterval

islinear

ismember

isnan

isreal

isrelaxfeasible

issigmonial

kkt

length

lifted

linearize

lmi

lmi2sedumistruct

lmiinfo

lmior

loadobj

logic2cont

lowrank

minus

or

plot

plotlattice

plus

polytope

problemclass

projection

pwamodel

reduce

remap

replace

saveobj

sdpvar

see

set

setcutflag

setdualize

settype

setupMeta

shift

size

sizeOLD

sosd

spy

subsasgn

subsref

tag

uncertain

uplus

usedvariables

value

variablereplace

vertcat

Error using optimize

Too many input arguments.

Error in sdpex (line 7)

optimize(F,[],sdpsettings('debug',1))

Johan Löfberg

unread,

Jul 20, 2017, 6:49:27 AM7/20/17

to YALMIP

For some bizarre reason, the wrong optimize method is called (optproblem method even when you don't use optprolem objects). Something is broken

Edit optproblem/optimize and add this before the if-statement, run sdpex and report what is displayed

dbstack

Johan Löfberg

unread,

Jul 20, 2017, 6:50:37 AM7/20/17

to YALMIP

and run yalmiptest and report the answer

Can Li

unread,

Jul 20, 2017, 7:37:21 AM7/20/17

to YALMIP

Can Li

unread,

Jul 20, 2017, 7:40:42 AM7/20/17

to YALMIP

The same errors reported. Should i reinstall the YALMIP?

Johan Löfberg

unread,

Jul 20, 2017, 8:01:39 AM7/20/17

to YALMIP

Remove the directory extras/@optproblem

Johan Löfberg

unread,

Jul 20, 2017, 8:02:01 AM7/20/17

to YALMIP

That is the first thing you should have tried. I assumed you already had try to delete everything and installed from scratch

Can Li

unread,

Jul 20, 2017, 8:42:38 AM7/20/17

to YALMIP

Yes! The problem is now solved by reinstalling the YALMIP.

Thank you so much!

Johan Löfberg

unread,

Jul 20, 2017, 9:07:14 AM7/20/17

to YALMIP

OK, great

It would have been interesting to have understood what caused this issue, very mysterious, but too late now

Can Li

unread,

Jul 20, 2017, 10:30:02 AM7/20/17

to YALMIP

May be something lost when the first installation, because I install and use this tool just yesterday. I choose YALMIP for its ability to process equality constraints and it seems more convenient to solve LMIs than the traditional one.

I have another question that how can i transform an Element-wise inequality to a Linear matrix one? For example, using the traditional LMI tool, one can realize it by writhing 0.5*(A'+A)<0 to replace A<0 (matrix A may not be symmetric).

I have tried it in YALMIP but failed.

Johan Löfberg

unread,

Jul 20, 2017, 1:06:37 PM7/20/17

to YALMIP

If you have a non-symmetric matrix A and you want it to have non-negative values, you write A<=0. If you want the unsymmetric matrix A to satisfy v'Av<=0 for all v, then you have to symmetrize it and write A+A' <= 0. If you want the unsymmetric matrix A to have non-positive eigenvalues, you jhave problems because that is a very hard nonconvex constraint and not an LMI.

Reply all

Reply to author

Forward