Nick Sahinidis recognized that the models nuclear14 and nuclear24,
nuclear14a and nuclear24a, and nuclear14b and nuclear24b
are equal, respectively. As a consequence, the nuclear24*
models have been removed from the MINLPLib.
Replaced infeasible point chp_partload_p2 by chp_partload_p3 and
infeasible points tln6_p2 and tln7_p2 by feasible ones.
Added $offlisting to the top of all model files to
allow for shorter listing files.
Nick Sahinidis recognized that the models detf1 and saa_2 are
equal. As a consequence, model detf1 has been removed
from the MINLPLib. The new best known solution for saa_2
Updated MINLPLib statistics to correctly reflect best known value
for water4 (907.017) as updated on 11/05/11.
Removed solution points chp_partload_p4.gdx and contvar_p2.gdx,
since they were not feasible. This increases the best
known value for contvar to 809149.8272.
New improved solutions for models eg_all_s (7.6578), mbtd(4.8333),
saa_2 (12.6922), water3 (907.017), waters (913.776),
watersym2 (940.9189), and watersbp (913.7757).
The points have been found by SCIP.
New improved solutions for models saa_2 (12.7375), tls7 (15.5),
waters (935.6923), and watersbp (947.2880). The points have
been found by SCIP. The new solutions for waters, watersbp
and tls7 replace the ones added on 11/05/11.
New improved solutions for models tls5 (10.3), tls6 (15.3), tls7 (15.8),
nuclear25b (-1.1010), water3 (920.6722), water4 (907.017),
watersbp (961.2847), waters (938.9031), watersym1 (913.7757)
have been added. These points have been found by SCIP.
CS Norman Shiau contributed 3 new models that perform an optimal vehicle
allocation for minimizing greenhouse gas emissions:
ghg_1veh, ghg_2veh, ghg_3veh
Toni Lastusilta pointed out that for some of the fo7, mo7, and o7 instances,
some provided points are only feasible w.r.t. very loose tolerance,
e.g., 1e-3. As a consequence, the points fo7_p1, fo8_p1, fo9_p1,
o7_2_p1, and o7_p1 have been removed and the points fo7_2_p1 and
mo7_p1 have been replaced by more accurate ones. Note that this
increases the best known value for fo7_2 to 17.7493 (previously
Toni Lastusilta contributed improved solutions for models
nuclear14b (-1.1169), nuclear24b (-1.1169), nuclear49b (-1.1169),
nuclearvd (-1.0344), tls12 (112.8), water3 (930.9874),
waterful2 (1012.6093), watersym1 (941.5144), waterx (909.0401).
The solutions were found via AlphaECP.
Pietro Belotti contributed an improved solution with value -115656.5
for model cecil_13.gms. He ran Couenne/trunk for 21000 seconds,
enumerating 2.1 million branch-and-bound nodes.
lop97ic.gms: A solution of value 4041.83 has been found by
LindoGlobal. The solution is recorded in point
Giacomo Nannicini contributed global solutions for the models water4
and waterz. He used Couenne with strong branching and a new
aggressive probing bound tightening technique.
Stefan Vigerske kindly agreed to become the next managing editor of
MINLPWorld. Thanks to Michael Bussieck for more than eight
years of service. Michael remains with MINLPWorld as a new
member of the advisory board.
Tapio Westerlund and Toni Lastusilta contributed four new models:
netmod_dol1,_dol2,_kar1, and _kar2 with points.
Tapio Westerlund and Toni Lastusilta contributed various new points
and pointed out some bugs in the points file. Jon Lee (IBM)
contributed the global solution to tln12. Stefan Vigerske
contributed the global optimum solution to lop97icx. Klaus
Schittkowski and Thomas Lehmann reported a new solution for
Stefan Vigerske suggested to change the quadratic term **2
to sqr() in model product and product2.
Tapio Westerlund and Toni Lastusilta contributed improved points for
the models fo7, fo8, fo9, o7, o7_2, tln12, and waters using
AlphaECP. SBB when restarted from the AlphaECP solution
found an even better point for tln12 (p2). The incorrect
point p1 for nvs14 has updated.
Stefan Vigerske contributed a new model chp_partload related to the
super* models already in the library. He also provide four
points found by SBB, LaGO, and OQNLP.
Reorganization of the point files: The ASCII point files have been
changed into GDX files for easier use. The GAMS command
'execute_loadpoint "m_p1.gdx"' reads the point into GAMS. A
wrapper called loadpoint has been added to the library, so
loading of points from the command line can be used as
gams batch u1=loadpoint --pfile batch_p1.gdx
In case you liked the old point format. You can use the GDX
utility gdxdump to generate ASCII based point format:
'gdxdump batch_p1.gdx format=gamsbas > batch.p1' and then
run the model: 'gams batch u1=batch.p1'.
Tapio Westerlund contributed two models: csched1a and csched2a.
These models are the corrected ones from the paper. The
printed version of the paper had some data inconsistencies.
The objective of the models also has been reformulated.
Points for these models also available.
Monique Guignard contributed instances from the generalized quadratic
assignment problem: pb302035, pb302055, pb302075, pb302095,
pb351535, pb351555, pb351575, pb351595. AO and ECP methods
do very well on those. Points have been given for all of
Toni Lastusilta added bounds to some existing models to avoid
evaluation errors with using the AlphaECP solver. The new
models are: enpro48pb, enpro56pb, ex3pb, ravempb, and
risk2bpb. Points have been given for all of them.
New points: There were a few models without an integer solution. We
added some solutions using various solvers: waterful2_p2,
m7_p1, m6_p1, m3_p1, fo7_2_p1, fo7_p1, fo8_p1, fo9_p1,
o7_2_p1, o7_p1, tls5, tls6.
The only models in MINLPLib without an integer solution are
nuclear104, nuclear10a, nuclear10b, space960 and
The listing of the best known solution on the model page contains now
a comment. This might indicate that the solution is the
global optimum solution. We give the reason why we think
this is the global optimum. In most cases, a global solver
like BARON provided the solution. Please note that all
solvers work with tolerances, so globally optimal is subject
to the solver tolerances (optimality as well as
Starting with GAMS22.5 there is a new global solver called LINDOGLOBAL
from Lindo Systems, Inc. LINDOGLOBAL accepts models with
trigonometric functions (sin, cos, ...). With that we have
been able to mark some more solutions to problems as
There are 103 quadratic problems among the 260 models in MINLPLib. The
model minlplib.gms contains the set m_miqcp with these
models. The scalar models have been adjusted to the model
type on the solve statement can be adjusted. This is also
useful, if one wants to solve the relaxed model. Just add
--minlp=miqcp or --minlp=rminlp to your model line to set
the model type in the solve statement.
Stefan Vigerske provided new solutions for models pump (p2) and
eq_int_s (p1), Both solutions were found with LaGO/Conopt
and verified using SBB and CONOPT.
Stefan Vigerske provided new solutions (p2) for models nuclear14b,
nuclear24b, and nuclear25n. Both solutions were found with
LaGO and verified using SBB and CONOPT.
These points have been improved by SBB/CONOPT (p3). The
tryint option in SBB lets one reproduce a given point by
setting branching directions on discrete variables. The SBB
option "dfsstay" does a couple of node explorations in the
neigborhood of the integer solution found and find the
"best" solution within a neighborhood. Using this approach
all three points have been slightly improved.
Tapio Westerlund also provided solutions for the models added on 12/19/05.
All solutions were found with AlphaECP and verified using SBB
32 new models fo[7,8,9]_ar*, m7_ar*, no7_ar* o[7,8,9]_ar* added: Tapio
Westerlund contributed these model. They comes from the
paper 'Castillo, I, Westerlund, J, Emetb, S, and Westerlund,
T, Optimization of block layout design problems with unequal
areas: A comparison of MILP and MINLP optimization methods.
Computers and Chemical Engineering 30 (2005), 54-769.
Two new GAMS Client models added: uselinear and blendgap. Furthermore,
Stefan Vigerske contributed four model (super1,2,3,and 3t)
together with points generated by LaGO.
New improved point for tls4. The solution was obtained by Nick
Sahinidis and Mohit Tawarmalani using BARON in the paper
"A polyhedral branch-and-cut approach to global
optimization" to appear in Math. Prog.
Facility layout problems fo7_2, fo7, fo8, fo9, m3, m6, m7, o7_2, and
o7 added. Tapio Westerlund contributed these. The original
models are part of the AlphaECP package. These are difficult
New improved points for def1, dosemin3d, eniplac, fuzzy, hda and
waterz. The solutions were obtained by AOA, the AIMMS Outer
Approximation algorithm. The points were contributed by
tls4.gms: A solution of value 9.30 has been found by by a novel
simplicial-approximation based algorithm (Goyal and
Ierapetritou). The solution is recorded in point tls4.p2.
mbtd.gms added: A GAMS Development client model.
prob02.gms, prob03.gms, prob10.gms added: Example models from the AlphaECP
system by T. Westerlund and K. Lundqvist.
Updated URL for MacMINLP.
nvs01.gms-nvs24.gms added: A collection of small non-convex models from a 1985
paper by Gupta and Ravindran. The global optimum solution is known
for these problems. Contributed by Nick Sahinidis.
product.gms and product2.gms: Two product positioning models contributed by Aldo
fuzzy.gms and risk2b.gms: Two GAMS Development client models.
du-opt5.gms, du-opt.gms, eg_all_s.gms eg_disc_s.gms eg_disc2_s.gms eg_int_s.gms:
Examples from Bram Schoonen's Master Thesis 'Mixed Integer Non-linear
Programming in Design Optimization"
For all new models feasible (for nvs* problems even global optimal) points have
been found and added to the library.
windfac.gms, elf.gms, sep1.gms, pump.gms: Added initial values.
lop97ic.gms: A solution of value 4284.59 has been found by
SBB/CONOPT3. The solution is recorded in point
nous1.gms N. Sahinidis reported an error in this model which caused
the infeasibility of the model. The RHS of equation e27
reads now 50 instead of 70. A global optimum point is also
added to the library.
lop97icx.gms and lop97ic.gms added. The first one is derived from the
latter by fixing variables that were zero in the relaxed problem. A
solution of lop97icx.gms can be transfered into a solution of lop97ic.
The references for the models have been redone.
For models where the original source is available we have a link to
tln6.gms The global solution of value 15.3 has been found by
BARON. The solution is recorded in point tln6.p2. This
updates the best know solution of value 15.6 which is
recorded in tln6.p1
tln7.gms The global solution of value 15.0 has been found by
BARON. The solution is recorded in point tln7.p2. This
updates the best know solution of value 19.5 which is
recorded in tln7.p1
waste.gms: A solution of value 598.919176 has been found by
SBB/SNOPT/CONOPT. The solution is recorded in point
waste.p2. This updates the best know solution of value
696.69 which is recorded in waste.p1
MINLPLib released with 136 model from 43 applications with 130 points.
126 of the models have some best known solution. For 10 models we do
not have an integer solution including 4 models for which we can't
provide a solution of the relaxation.