[Ipopt] solver MA27 out of memory

[Tony Kelman]

Is Ipopt failing at that iteration 76, or does it continue? What’s more important than the number of variables is the number of nonzeros in the Hessian and Jacobian. Some problems of that size MA27 should be fine, but if your problem is very dense or numerically challenging then it may start having trouble. Most times I’ve seen that flag it has just been a warning, a diagnostic message that your problem is resulting in a large amount of “fill-in” during the factorization. If you have academic access to the other solvers, it’s worth giving all of them a try for comparison purposes. You just have to have them available when you compile Ipopt, or in an external library and use the linear solver loader.

-Tony


From: leihao 
Sent: Monday, October 28, 2013 3:06 AM
To: ipopt 
Subject: [Ipopt] solver MA27 out of memory

hi all:

I use  MA27 as my default solver.
There are 15000 Variables for my question .
the input data  may be  too large and ipopt print  information as below:

  66r 2.7573601e+12 1.18e+06 9.88e+02   6.9 2.49e+08    -  5.62e-05 7.61e-03f  1
  67r 2.7545993e+12 1.18e+06 9.88e+02  -0.1 1.73e+10    -  9.70e-05 1.12e-03f  1
  68  4.7698193e+11 4.40e+05 5.37e-01   2.2 3.33e+07    -  1.12e-02 6.26e-01h  1
  69  1.8904455e+11 1.13e+05 1.09e+00   2.4 2.81e+06    -  6.92e-02 7.43e-01h  1
iter    objective    inf_pr   inf_du lg(mu)  ||d||  lg(rg) alpha_du alpha_pr  ls
  70  3.0346243e+10 2.78e+04 6.92e-01  -1.9 7.51e+06    -  2.32e-01 7.54e-01h  1
  71  2.2450431e+09 3.56e+03 4.34e-01  -2.0 3.62e+06    -  2.58e-01 8.72e-01h  1
  72  4.0989559e+08 3.16e+03 5.04e+00   1.7 9.66e+05    -  7.13e-01 1.11e-01f  1
  73  2.1918911e+07 3.24e+02 5.18e-01  -2.7 8.67e+05    -  9.19e-01 8.97e-01h  1
  74  1.7622191e+07 4.05e+00 6.65e-03  -3.8 1.48e+05    -  9.78e-01 9.88e-01h  1
  75  1.7618256e+07 2.33e-09 1.44e-02  -5.4 8.50e+03    -  9.90e-01 1.00e+00h  1
MA27BD returned iflag=-4 and requires more memory.
Increase liw from 11075700 to 22151400 and la from 11620400 to 48600530 and factorize again.
  76  1.7337765e+07 2.33e-09 2.23e-01  -7.4 5.68e+03    -  9.61e-01 1.00e+00h  1
---
my ipopt.opt file:
ma27_liw_init_factor 100.0
ma27_la_init_factor 100.0
ma27_meminc_factor 2.0
constr_mult_init_max 0
nlp_scaling_method gradient-based
linear_solver ma27

---
Is the reason too much var ?
Can HSL_MA86 solve my problem?

I find the table from http://www.hsl.rl.ac.uk/ipopt/  :





-- 
 
裴垒浩
新浪微博-商业产品部
Tel：6061 9679
Mobile：186 1068 2796
E-mail：leihao at staff.sina.com.cn
微博：@红萝卜的帽子 http://weibo.com/iphxer
地址：北京市 海淀区 彩和坊路 朔黄发展大厦 10层

--------------------------------------------------------------------------------
_______________________________________________
Ipopt mailing list
Ipopt at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/ipopt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20131028/eb2ac9d4/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ebajifjb.png
Type: image/png
Size: 5108 bytes
Desc: not available
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20131028/eb2ac9d4/attachment.png>