[Help-glpk] glpk and MIPLIB 2003

[Cheng,Jen-Min]

MIPLIB 2003  http://miplib.zib.de/

I compiled the glpk 4.31 with Microsoft Visual Studio 2005. 

Glpk can solve manna81.mps of MIPLIB 2003 with –mipgap 1.0 in about 37 seconds.  –mipgap 0.05 ended with tree is empty after about 129138 seconds.

The two examples make me wonder how an appropriate mipgap can be selected.  In the first example, --mipgap 1.0 works.  In the second exmple, --mipgap 0.05 could not converge to the psoted solution.

Will you please provide guidelines for selecting mipgap.  Thanks.

manna81.mps

MIPLIB 2003

http://miplib.zib.de/

INT OBJ = -13164   LP OBJ=-13297

6480 Rows, 3321 Columns, 12960 Non-zeros

3303 Integers 18 Binaries

--mipgap 1.0

+  5184: mip = -1.316400000e+004 >= -1.329700000e+004   1.0% (285; 0)

Time used:   36.6 secs

However glpk with –mipgap 0.1 could not converge to the solution posted by MIPLIB 2003. 

vpm2.mps

MIPLIB 2003

http://miplib.zib.de/

INT OBJ = 13.75   LP OBJ=9.8892646

234 Rows, 378 Columns, 1085 Nonzeros

168 Binaries, 210 Continuos variables

--mipgap 1.0

+  2369: mip =  1.700000000e+001 >=  1.070274874e+001  37.0% (502; 45)

Time used:   0.7 secs

--mipgap 0.5

+  2369: mip =  1.700000000e+001 >=  1.070274874e+001  37.0% (502; 45)

Time used:   0.7 secs

--mipgap 0.3

+ 29961: mip =  1.525000000e+001 >=  1.156433411e+001  24.2% (6352; 680)

Time used:   10.2 secs

--mipgap 0.2

+ 33900: mip =  1.450000000e+001 >=  1.161038594e+001  19.9% (6720; 1581)

Time used:   9.5 secs

--mipgap 0.1

+1462797: mip =  1.375000000e+001 >=  1.282598934e+001   6.7% (257981; 57219)

Time used:   2630.4 secs

--mipgap 0.05

+1462797: >>>>>  1.375000000e+001 >=  1.282598934e+001   6.7% (257981; 57219)

+6072379: mip =  1.375000000e+001 >=  1.373563363e+001   0.1% (2431; +6074797: mip =  1.375000000e+001 >=     tree is empty   0.0% (0; 1940503)

Time used:   129138.0 secs