===firstname:
David
===firstname3:
Tom
===affil6:

===lastname3:
Manteuffel
===email:
dappelh@us.ibm.com
===keyword_other2:

===lastname6:

===affil5:

===lastname4:
McCormick
===lastname7:

===affil7:

===postal:
99 Grant Ave #1
Medford, MA 02155
===ABSTRACT:
Combining nested iteration, adaptive mesh refinement, and range decomposition creates a powerful PDE solution method with reduced communication costs compared to traditional solution methods. Range decomposition partitions a PDE according to the right side of the equation, rather than the domain. If the right side is approximately equally distributed, for example from a coarse adaptive least squares solve, then range decomposition equally distributes the error and is a naturally load balanced algorithm. This algorithm is naturally suited for large distributed memory systems where intra-node communication is expensive but on node computational power is plentiful. The method will be explained, a performance model will show the scaling benefits, and numerical results for an advection-diffusion type PDE will be shown. 
===affil3:
University of Colorado, Boulder
===title:
Nested Iteration Range Decomposition: A scalable, low communication PDE solution method with numerical results on advection-diffusion problem.
===affil2:
University of Colorado, Boulder
===lastname2:
Ruge
===firstname4:
Steve
===keyword1:
Iterative solvers/linear algebra on high concurrency node architectures
===workshop:
no
===lastname:
Appelhans
===firstname5:

===keyword2:
NOT_SPECIFIED
===otherauths:

===affil4:
University of Colorado, Boulder
===competition:
no
===firstname7:

===firstname6:

===keyword_other1:

===lastname5:

===affilother:

===firstname2:
John