.. _method-snowpac-trust_region-initial_size:

""""""""""""
initial_size
""""""""""""


Trust region initial size (relative to bounds)


.. toctree::
   :hidden:
   :maxdepth: 1



**Specification**

- *Alias:* None

- *Arguments:* REALLIST


**Description**


The ``trust_region`` optional group specification can be used to
specify the initial size of the trust region (using ``initial_size``)
relative to the total variable bounds, the minimum size of the trust
region (using ``minimum_size``), the contraction factor for the trust
region size (using ``contraction_factor``) used when the surrogate
model is performing poorly, and the expansion factor for the trust
region size (using ``expansion_factor``) used when the the surrogate
model is performing well. Two additional commands are the trust region
size contraction threshold (using ``contract_threshold``) and the trust
region size expansion threshold (using ``expand_threshold``).  These
two commands are related to what is called the trust region ratio,
which is the actual decrease in the truth model divided by the
predicted decrease in the truth model in the current trust region. The
command ``contract_threshold`` sets the minimum acceptable value for
the trust region ratio, i.e., values below this threshold cause the
trust region to shrink for the next SBL iteration. The command
``expand_threshold`` determines the trust region value above which the
trust region will expand for the next SBL iteration.

All of these specifications are ``REAL`` scalars, with the exception of
the specification of ``initial_size``, which is a ``REALLIST``.  This
array corresponds to the case when there are more than 2 model forms
or discretizations within a model hierarchy.  The default
``initial_size`` involves a recursive halving of the global bounds for
each trust region in the hierarchy: e.g., a scalar value of .5 for a
single trust region managing two model forms/discretizations, or an
array of ( .125, .25, .5 ) for four model forms/discretizations
(three trust regions ordered from the lowest to highest fidelity
surrogate, with model four as truth).