[ADMB Users] eigenvectors of the Hessian and parameter errors

[dave fournier]

On 11-10-07 11:42 AM, David Holmgren wrote:


Well if you want to send me what I need to run the model, I'll take a 
quick look at it.


> Thanks.  The model formulation could indeed be the problem.  I should also say that the four negative eigenvalues are all small, i.e., on the order of -1e-08 or less.
>
> Dave H.
>
> On 2011-10-07, at 12:20 AM, dave fournier wrote:
>
>> On 11-10-06 09:01 PM, David Holmgren wrote:
>>
>> OK,
>>
>> In that case you have a problem. So either you are not at a minimum or the way
>> the model is formulated it can not get quite close enough to make the Hessian positive definite.
>>
>>           Dave
>>
>>> Many thanks!  It is usually the case that four of 24 eigenvalues are negative.  I'm not fixing any parameters, or doing the minimization in phases.
>>>
>>> Dave H.
>>>
>>> On 2011-10-06, at 8:27 AM, dave fournier wrote:
>>>
>>>> I sent this once but it seems to have disappeared .. so again
>>>>
>>>> The error gets printed out in two cases I think.  1.) if the eigenvalues of the Hessian are not positive.
>>>> 2) If the  std dev of a dependent variables computed by the delta method is not positive. However
>>>> if that parameter is not active then the message will get produced since its gradient is 0.
>>>> There was some attempt to make this a bit less confusing at some time, but maybe it died somewhere.
>>>> _______________________________________________
>>>> Users mailing list
>>>> Users at admb-project.org
>>>> http://lists.admb-project.org/mailman/listinfo/users
>>> David Holmgren
>>> holmgren at telus.net
>>>
>>>
>>>
>>>
> David Holmgren
> holmgren at telus.net
>
>
>
>