maximize two objective function

Dear Gams experts,

Is it possible to maximize two objective functions?
For example, in the differential game in Economics (e.g. Nash or Stackelberg game), there are two objective functions.
Without the first order equations, is it possible to solve the differential game?

Thank you very much in advance.

Sincerely,

Q


To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

\

It depends completely on the behavior of players/objectives.
For instance if the idea is to compute Pareto solutions, i.e., cooperation
then your problem reduces to an optimal control problem and one can use GAMS routines like NLP etc…

If the idea is to compute Nash type solution, complementarity framework can be used to compute the equilibria from necessary conditions.
If you want to avoid necessary conditions, there is a way to solve such a game in an approximate way - iterative method/best reply method.
You could randomly initialize the strategies of players other than the player say ‘i’ and find the best response given by i,
you could update the i’s strategy as above and move to player i+1 fixing the others and ‘i+1’ gives a best response…
you could iterate this procedure till you get some convergence, which may or may not be guranteed. however, if the objectives are sufficiently convex
there is some hope.

However, while working with differential game one has to be careful about information structure, as in, open loop (just uses information about time and initial conditions) or closed loop (use information of history of state information) type. (refer Basar and Oldsder’s book on Dynamic Non Cooperative Game Theory’ more details).

hope this helps.

-Vishwa


On Saturday, July 27, 2013 1:28:33 AM UTC-4, environ...@gmail.com wrote:

Dear Gams experts,

Is it possible to maximize two objective functions?
For example, in the differential game in Economics (e.g. Nash or Stackelberg game), there are two objective functions.
Without the first order equations, is it possible to solve the differential game?

Thank you very much in advance.

Sincerely,

Q


To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

\

My apologies, Q, I should have looked into and thought about this more before providing only a terse response.

The Nash games are iterative, that is one player responds to the other player’s actions. Within GAMS, you could sequentially optimize separate objective functions but a simultaneous optimization doesn’t immediately seem to be conceptually possible given the recursive nature of the individuals’ decision variables. I will defer to the rest of the GAMS experts and look forward to seeing their responses.

Best,
Max

On Sunday, July 28, 2013 6:50:11 PM UTC-4, Maxwell Brown wrote:

Q,

If the objective functions’ variables are independent, they can just be combined into a single objective function.

If not, I would need more specifics regarding your intentions.

Best,
Max


To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

\

Dear Max,

For example, there are two players, and they want to maximize their own utility function. All variables include time variables.

Max u1(x1, y1)
s.t. xdot1= y1 + w…this is equations of motion
w = y1 if y1 > y2
otherwise, w = y2

Max u2(x2, y2)
s.t. xdot1 = y2 + w
w = y1 if y1 > y2
otherwise, w = y2

w function is connected between two players. In this case, I think 2 solve statements are needed.
What do you think?

Sincerely yours,

Q


On Saturday, July 27, 2013 1:28:33 AM UTC-4, environ...@gmail.com wrote:

Dear Gams experts,

Is it possible to maximize two objective functions?
For example, in the differential game in Economics (e.g. Nash or Stackelberg game), there are two objective functions.
Without the first order equations, is it possible to solve the differential game?

Thank you very much in advance.

Sincerely,

Q


To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

\

Dear Max,

Thank you very much for your response.

For example, there are two players, and they want to maximize their own utility function.

Max u1(x1, y1)
s.t. x1 = y1 + w
w = y1 if y1>y2
w = y2 otherwise

Max u2(x2, y2)
s.t. x2 = y2 + w
w = y1 if y1>y2
w = y2 otherwise

In this case, w function is connected between two players.
I think 2 solve statement are needed. What do you think?

Sincerely,

Q



On Saturday, July 27, 2013 1:28:33 AM UTC-4, environ...@gmail.com wrote:

Dear Gams experts,

Is it possible to maximize two objective functions?
For example, in the differential game in Economics (e.g. Nash or Stackelberg game), there are two objective functions.
Without the first order equations, is it possible to solve the differential game?

Thank you very much in advance.

Sincerely,

Q


To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

\

Dear Vishwa and Max,

Thank you very much for your response.

My problem is the differential game and open loop.

Max u1(x1, y1)
s.t. xdot1= y1 + w…this is equations of motion
w = y1 if y1 > y2
otherwise, w = y2

Max u2(x2, y2)
s.t. xdot1 = y2 + w
w = y1 if y1 > y2
otherwise, w = y2

When I coded this problem,

Equations;

x1… x1(t+1) = x1(t) + y1(t) + w(t)
x2… x2(t+1) = x2(t) + y2(t) + w(t)
w… w(t) = y1(t)…This is binary equation.
objective 1… U1 = u1(t)
objective 2… U2 = u2(t)

Model Nash /all/;
Solve Nash maximizing U1 using MINLP;
Solve Nash maximizing U2 using MINLP;

I added two solve statement to solve this problem simultaneously, but I would like to know if this problem is a non-cooperative differential game and open loop, then is it correct to code this problem?

Thank you very much in advance.

Sincerely yours,

Q




Do you have any references about coding in GAMS?





On Saturday, July 27, 2013 1:28:33 AM UTC-4, environ...@gmail.com wrote:

Dear Gams experts,

Is it possible to maximize two objective functions?
For example, in the differential game in Economics (e.g. Nash or Stackelberg game), there are two objective functions.
Without the first order equations, is it possible to solve the differential game?

Thank you very much in advance.

Sincerely,

Q


To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

\

There is an example of a Nash/Stackleberg in the GAMS model library, if you don’t know where this is:
GAMS IDE → Model Library → Nash

Vishwa has great insight into these games and I find this line to be your best route:
“you could update the i’s strategy as above and move to player i+1 fixing the others and ‘i+1’ gives a best response…”

I believe you might want to get rid of time subscripts here, each set of solves would be an iteration through T – Vishwa do you have thoughts on this? It is also possible, as shown by the example in the model library, to simulate the game in a gen eq framework.

With respect to your code:

x1… x1(t+1) = x1(t) + y1(t) + w(t)
x2… x2(t+1) = x2(t) + y2(t) + w(t)
w… w(t) = y1(t)…This is binary equation.
objective 1… U1 = u1(t)
objective 2… U2 = u2(t)

Model Nash /all/;
Solve Nash maximizing U1 using MINLP;

#Record the move (i.e. Y) by player 1:
y1t = y1.L(t);

#Notice highlighting
xdot1 = y2 + w
w = y1t if y1t > y2
otherwise, w = y2

Solve Nash maximizing U2 using MINLP;

#Record the move (i.e. Y) by player 1:
y2t = y2.L(t);

xdot1 = y1 + w
w = y2t if y2t > y1
otherwise, w = y1

#This is an example why the T indices would need to be controlled differently, you could create a set called ‘currT’

Solve Nash maximizing U1 using MINLP;


As for a GAMS reference, McCarl is where I started and a very thorough, well-written reference:
http://www.gams.com/dd/docs/bigdocs/gams2002/mccarlgamsuserguide.pdf

I hope this helps - the example in the model library might cause you to rework your problem unless you like the iterative approach.

Best,
Max

On Monday, July 29, 2013 5:34:06 AM UTC-4, environ...@gmail.com wrote:

Dear Vishwa and Max,

Thank you very much for your response.

My problem is the differential game and open loop.

Max u1(x1, y1)
s.t. xdot1= y1 + w…this is equations of motion
w = y1 if y1 > y2
otherwise, w = y2

Max u2(x2, y2)
s.t. xdot1 = y2 + w
w = y1 if y1 > y2
otherwise, w = y2

When I coded this problem,

Equations;

x1… x1(t+1) = x1(t) + y1(t) + w(t)
x2… x2(t+1) = x2(t) + y2(t) + w(t)
w… w(t) = y1(t)…This is binary equation.
objective 1… U1 = u1(t)
objective 2… U2 = u2(t)

Model Nash /all/;
Solve Nash maximizing U1 using MINLP;
Solve Nash maximizing U2 using MINLP;

I added two solve statement to solve this problem simultaneously, but I would like to know if this problem is a non-cooperative differential game and open loop, then is it correct to code this problem?

Thank you very much in advance.

Sincerely yours,

Q




Do you have any references about coding in GAMS?





On Saturday, July 27, 2013 1:28:33 AM UTC-4, environ...@gmail.com wrote:

Dear Gams experts,

Is it possible to maximize two objective functions?
For example, in the differential game in Economics (e.g. Nash or Stackelberg game), there are two objective functions.
Without the first order equations, is it possible to solve the differential game?

Thank you very much in advance.

Sincerely,

Q


To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at http://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/groups/opt_out.

\