Introduction
============

The objective of the GridPACK\ :math:`\mathrm{{}^{TM}}` toolkit project
is to provide a framework to support the rapid development of power grid
applications capable of running on high performance computing
architectures (HPC) with high levels of performance and scalability. The
toolkit allows power system engineers to focus on developing working
applications from their models without getting bogged down in the
details of decomposing the computation across multiple processors,
managing data transfers between processors, working out index
transformations between power grid networks and the matrices generated
by different power grid applications, and managing input and output.
GridPACK is being designed to encapsulate as much of the book-keeping
required to set up HPC applications as possible using high-level
programming abstractions that allow developers to concentrate on the
physics and mathematics of their problems. This document summarizes the
overall design of the GridPACK framework and provides a detailed
description of its components. The remainder of this document will
describe the functionality incorporated into the GridPACK framework to
support multiple power grid applications. The framework will continue to
evolve as more real-world experience can be incorporated into the design
process but many base classes that have already been identified that are
capable of supporting a range of applications. During the initial stages
of GridPACK development, four power grid applications were targeted for
implementation. These included:

#. Powerflow simulations of the electric grid

#. Contingency analysis of the electric grid

#. State estimation based on electric grid measurements

#. Dynamic simulations of the electric grid

From these applications, several cross-cutting functionalities were
identified that could be used to support multiple applications.

#. Network topology and behavior. The network topology is the starting
   point for any power grid analysis. The topology defines the initial
   network model and is the connection point between the physical
   problem definition in terms of buses and branches and the solution
   method, which is usually expressed in terms of matrices and vectors.

#. Network components and their properties (e.g. bus and branch models,
   measurements, etc.). Grid components are the objects associated with
   the buses and branches of the power grid network. Along with the
   network topology itself, these define the physical system being
   modeled and in some cases the analysis that is to be performed. Bus
   and branch components can be differentiated into things like
   generators, loads, grounds, lines, transformers, measurements, etc.
   and depending on the how they are defined and the level of detail
   incorporated into them, they define different power grid systems and
   analyses. The behavior of buses and branches can depend on the
   properties of branches or buses that are directly attached to them,
   e.g. figuring out the contribution of a particular bus to the
   solution procedure may require that properties of the attached
   branches are made available to the bus. The capability for exchanging
   this data is built into the framework. Furthermore, these data
   exchanges must also be accounted for in a parallel computing context,
   since the grid component from which data is required may be located
   on a different processor.

#. Linear algebra and solvers. Basic algebraic objects, such as
   distributed matrices and vectors, are a core part of the solution
   algorithms required by power grid analyses. Most solution algorithms
   are dominated by sparse matrices but a few, such as Kalman filter
   analyses, require dense matrices. Vectors are typically dense. There
   exists a rich set of libraries for constructing distributed matrices
   and vectors and these are coupled to preconditioner and solver
   libraries. GridPACK can leverage this work heavily by creating
   wrappers to these libraries that can be used in solution algorithms.
   Wrapping these libraries instead of using them directly will have the
   advantage that creating algebraic objects can be simplified somewhat
   for power grid applications but more importantly, it will allow
   framework developers to investigate new solver and algebraic
   libraries seamlessly, without disrupting other parts of the code.

#. Mapping between network and algebraic objects. The physical
   properties of power grid systems are defined by networks and the
   properties of the network components but the equations describing the
   networks are algebraic in nature. The mappings between the physical
   networks and the algebraic equations depend on the indexing scheme
   used to describe the network and the number of parameters in the
   network components that appear in the equations. Constructing a map
   between network parameters and their corresponding locations in a
   matrix or vector can be complicated and error prone. Fortunately,
   much of this work can be automated and developers can focus more on
   developing code to evaluate individual matrix elements without
   worrying about where to locate them in the matrix. This can simplify
   coding considerably.

The elements described above have all been incorporated into GridPACK
modules. More details about these modules and their interactions are
provided in the remainder of this document.