[Clp] Non-monotone behaviour of object value by primalSimplex

[vladimir voloshinov]

Dear CLP team,
1) Is it possible to ensure monotone change of object value during
iterations of "primalS" method?
(It looks like obj change monotonically during "dual simplex", after
"solve" command... Is it true?)
2) What is a "structure" of CLP log for "primalS" command (in brief)
or give some link to explanations,
http://www.coin-or.org/Clp/userguide/clpuserguide.html is oriented for
"CLP API programmers"...

Some reasons for the question:
we are dealing with rather large LP problems, e.g.
... 121035 rows, 9782472 columns and 29397353 elements ...
"solve" (Dual simplex) takes ~50 minutes to solve every instance and
it is too long...
There is an idea to get an approximate solution (feasible in primal
problem) by primalS with estimation of "un-optimality gap"  by
simultaneously solving (in parallel) of dual problem.

CLP log for "solve" command (dual simplex) corresponds my
"expectations" and obj value decreases (for maximization problem)
monotonically.

But after "primalS" for the same problem I saw (fragments):
27909  Obj 7.371526e+09 Primal inf 71024.993 (2002) Dual inf
4.9189008e+14 (2036138)
28268  Obj 7.401055e+09 Primal inf 70807.993 (1989) Dual inf
4.870662e+14 (1965642)
28717  Obj 7.4408544e+09 Primal inf 70192.993 (1893) Dual inf
4.8425552e+14 (2020039)
29280  Obj 7.4413182e+09 Primal inf 68595.993 (1831) Dual inf
6.1040781e+14 (1963375)
(increasing, for maximization, but it is already bigger than real
optimum ~ 5.52298E+09 !!??, may be it is still searching the first
primal feasible basis?)
...
57936  Obj 8.3215142e+09 Primal inf 45263.986 (515) Dual inf
9.2250956e+14 (2201843)
58671  Obj 8.3055334e+09 Primal inf 44212.986 (502) Dual inf
3.2978764e+14 (1941322)
59406  Obj 8.1377067e+09 Primal inf 43003.986 (486) Dual inf
3.2644637e+14 (1931661)
60143  Obj 8.0444598e+09 Primal inf 41654.986 (468) Dual inf
3.2437796e+14 (2000651)
60880  Obj 7.9114645e+09 Primal inf 40364.982 (454) Dual inf
9.9919246e+15 (3269701)
(decreasing !?)
...
82947  Obj 6.2828834e+08 Primal inf 992.97642 (86) Dual inf
2.8076612e+14 (2613853)
83684  Obj 6.2397507e+08 Primal inf 378.99121 (67) Dual inf
1.4655789e+16 (3398845)
84419  Obj 6.4070341e+08 Dual inf 2.7270238e+12 (4008734)    <= what a
strange "phase transition", a feasible basis has been found?
85154  Obj 8.7460819e+08 Dual inf 4.0494125e+11 (4258212)
85891  Obj 9.2782176e+08 Dual inf 3.590765e+11 (3987150)
86626  Obj 9.6103668e+08 Dual inf 2.3537481e+11 (3699157)
87362  Obj 1.0123088e+09 Dual inf 1.7949174e+11 (3978093)
(then Obj increases permanently ...)
...

Sincerely yours,
Vladimir Voloshinov,
Ph.D, head of lab. C-3 "Distributed computing algorithms"
Center of Grid-technologies & Distributed Computing, http://dcs.isa.ru,
web: http://dcs.isa.ru/drupal/ru/staff/vladimirv