[Bug] NullSpace[m, Method->OneStepRowReduction] can cause crashes in v5
- To: mathgroup at smc.vnet.net
- Subject: [mg49577] [Bug] NullSpace[m, Method->OneStepRowReduction] can cause crashes in v5
- From: Scott Morrison <scott at math.berkeley.edu>
- Date: Fri, 23 Jul 2004 06:01:49 -0400 (EDT)
- Reply-to: scott at math.berkeley.edu
- Sender: owner-wri-mathgroup at wolfram.com
Hi again,
I'm using $Version == 5.0 for Microsoft Windows (November 18, 2003).
For certain (large) matrices m, NullSpace[m,
Method->OneStepRowReduction] causes a bad kernel crash (the kernel dies,
and I get a Windows XP dialog about an error having occured, yada yada)
For example (sorry this is so large; it's a real life example...)
T = SparseArray[{{217, 7} -> 1, {217, 8} -> q, {310, 10} -> 1, {310, 11}
-> q, {403, 13} -> 1, {403, 14} -> q, {464, 15} -> 1, {464, 17} -> q,
{495, 16} -> 1, {495, 18} -> q, {589, 19} -> 1, {589, 20} -> q, {682,
22} -> 1, {682, 23} -> q, {775, 25} -> 1, {775, 26} -> q,
{1178, 5} -> 1, {1178, 7} -> q, {1207, 6} -> 1, {1207, 10} -> q,
{1300, 9} -> 1, {1300, 13} -> q, {1394, 12} -> 1,
{1394, 15} -> q, {1580, 18} -> 1, {1580, 21} -> q, {1641, 20} -> 1,
{1641, 24} -> q, {1734, 23} -> 1, {1734, 27} -> q,
{1829, 26} -> 1, {1829, 28} -> q, {2140, 3} -> 1, {2140, 5} -> q,
{2171, 4} -> 1, {2171, 6} -> q, {2449, 13} -> 1,
{2449, 16} -> q, {2479, 14} -> 1, {2479, 18} -> q, {2510, 15} -> 1,
{2510, 19} -> q, {2573, 17} -> 1, {2573, 20} -> q,
{2884, 27} -> 1, {2884, 29} -> q, {2915, 28} -> 1, {2915, 30} -> q,
{3134, 2} -> 1, {3134, 3} -> q, {3256, 6} -> 1,
{3256, 9} -> q, {3380, 10} -> 1, {3380, 13} -> q, {3411, 11} -> 1,
{3411, 14} -> q, {3659, 19} -> 1, {3659, 22} -> q,
{3690, 20} -> 1, {3690, 23} -> q, {3814, 24} -> 1, {3814, 27} -> q,
{4002, 30} -> 1, {4002, 31} -> q, {4189, 3} -> 1,
{4189, 4} -> q, {4251, 5} -> 1, {4251, 6} -> q, {4311, 7} -> 1, {4311,
10} -> q, {4342, 8} -> 1, {4342, 11} -> q,
{4776, 22} -> 1, {4776, 25} -> q, {4807, 23} -> 1, {4807, 26} -> q,
{4933, 27} -> 1, {4933, 28} -> q, {4995, 29} -> 1,
{4995, 30} -> q, {5151, 1} -> 1, {5151, 2} -> q, {5397, 9} -> 1,
{5397, 12} -> q, {5522, 13} -> 1, {5522, 15} -> q,
{5552, 14} -> 1, {5552, 17} -> q, {5614, 16} -> 1, {5614, 19} -> q,
{5677, 18} -> 1, {5677, 20} -> q, {5769, 21} -> 1,
{5769, 24} -> q, {6081, 31} -> 1, {6081, 32} -> q, {_, _} -> 0},{6144,32}]
On my computer at least, NullSpace[T, Method->OneStepRowReduction]
returns the correct answer, but then induces a kernel crash. (Often, but
not always, the kernel only crashes on the *next* input, for example
N[Pi,10^5]). Just NullSpace[T] returns the correct answer. Running the
same code on $Version == "4.2 for Microsoft Windows (June 5, 2002)" I
have no problems.
For now I'm just using NullSpace[T], but my experience while using 4.2
was that OneStepRowReduction worked much faster for the matrices I was
dealing with, so hopefully we'll hear about a fix for this! (Wolfram!?)
In the meantime, I'm offering this as (yet another?) bug in v5 which
took an inordinate amount of time to track down, and seems worth
watching out for.
Scott Morrison