Weblogic crash without a trace (core dump, hs_err_pid, stracktrace, logs)

Question

We have a weblogic 9 cluster where a node crash/shutdown with no trace. Sometimes its node 2 and sometimes node 3, but not at the same time.
We can't find any core dump, hs_err_pid or any stracktrace in the logs or std out/err.
Specifying where hs_err files goes with -XX:ErrorFile should not be necessary as it should write in current directory or if not possible due to permissions etc, in the OS tmp dir.
Are there any Weblogic specific JVM or system property we can specify to force some kind of trace when the process just seem to crash?
We see this in the log from time to time, perhaps its related:
<2012-jun-08 kl 13:53 CEST> <Error> <Security> <BEA-090060> <The AccessDecision class "weblogic.security.providers.realmadapter.AuthorizationProviderImpl" returned an error: java.lang.SecurityException: Realm Adapter ACL Mapping Failed.>
Can the above hide a possible out of memory error trace to appear in the logs?
SERVER: Weblogic 9.2.3 with some additional patches (and compatibility mode)
OS: HP-UX B.11.23 U ia64
JAVA: Java(TM) 2 Runtime Environment, Standard Edition (build 1.5.0.20-_28_apr_2010_03_15)
Java HotSpot(TM) Server VM (build 1.5.0.20 jinteg:04.28.10-02:28 IA64, mixed mode)

Answer 1

http://docs.oracle.com/cd/B28359_01/java.111/b31224/instclnt.htm gives some info about the used libraries (libocijdbc11.so)
Could you check the environment variable - http://docs.oracle.com/cd/B28359_01/java.111/b31224/getsta.htm#i1005378
"On Sun Solaris or Linux, set the LD_LIBRARY_PATH environment variable as follows:
ORACLE_HOME/lib
This directory contains the libocijdbc11.so shared object library.
Note:
If you are running a 32-bit Java Virtual Machine (JVM) against a 64-bit client or database, then you must also add ORACLE_HOME/lib32 to the LD_LIBRARY_PATH environment variable."