Mar 14 Notes

Rice’s and NCAR’s Internet II connection to the AG was erratic during the meeting. Dartmouth AG host Gurcharan Khanna suggested that all AG operators implement some software enhancements for future backup capabilities.

The 1^st annual CISM site visit will be held in Boston on May 29-30. The focus will be on CISM management structure. J Hughes noted that the monthly AG meeting of this MI WG serves to demonstrate that the distributed center concept is working well and that IP-based teleconferencing is facilitating progress in meeting CISM goals. J Hughes does not expect the 1^st site visit to focus on technical results so the MI WG will not need to prepare any research-related material for presentation at the site visit.

1) MI Coupling Flowchart: W Lotko presented a provisional flowchart for MI coupling linking the LFM, RCM and TING model. This flowchart provides an overview of the principal variables passed between codes for electrodynamic coupling, grid conversions required to do so and time cadence of the codes and coupling. A flowchart for inertial coupling involving mass exchange still needs to be developed. Although the introduction of a common Poisson solver for all 3 codes will introduce additional computational overhead and numerical diffusion, J Lyon noted that the Poisson solve step is actually a minor computational load in comparison with the fluid/MHD/drift calculations and can be implemented with a high-resolution grid to minimize numerical diffusion resulting from interpolations between model grids and the MIC Coupler grid.

One extension to the current electrodynamic coupling scheme (suggested by W Lotko) would include the contribution of the field-aligned current from LFM/RCM as a direct source term in the TING continuity equation for E-region plasma density. While the effects of FAC are already modeled in TING via a precipitation source term, the effects of density depletion in regions of downward FAC are not described by precipitation. S Solomon indicated that a number of more pressing model improvements are ahead of this one.

A revised flowchart with a legend slide added (per H Spence’s request) is available at the above referenced URL. Suggestions for improvements to the flowchart should be forwarded to W Lotko.

2) Grid Overlay Figure: The illustrative grid overlay figure developed for the 9/02 CISM kick-off meeting showing TIGCM, LFM and RCM grids at a reference ionospheric level was revisited. Evidently the RCM grid shown in the figure is oversimplified and could be embellished for future presentation.

3) Overture and MetaChaos: The LLNL Overture software package described by M Wiltberger was originally developed for solving systems of PDEs on multiple grids. It should also be adaptable to the CISM MI coupling problem, potentially simplifying exchange of variables between CISM (and other) codes running on different grids with different time steps. Grid attributes are defined and represented within Overture by a single reference module (e.g., mygrid.hdf), making implementation of transformation rules for passing data from one grid to another a relatively high-level coding procedure. However, if the grid can't be derived from one of Overture’s standard grid types, as is currently the case with the LFM and TING grids, then the grid points must be defined along with a few other pieces of useful information. Overture is also equipped with a toolkit that includes a finite-difference operations library, grid generation capabilities, an interface for PDE solution on adaptive and curvilinear grids, and support for serial and parallel operations. Calls to Fortran code from Overture’s C++ protocol and calls to Overture C++ routines from Fortran code are easily implemented. F Toffoletto questioned the performance of Overture’s general FD toolkit. J Lyon and M Wiltberger acknowledged that one would probably realize significant speedups (x2-3) with application-optimized code. However, J Lyon also noted that the matrix library used by Overture is highly optimized and may perform competitively for the Poisson solve step at the heart of the MI coupling problem.

MetaChaos is a runtime meta-library that achieves direct data transfers between data structures managed by different parallel libraries (MxN problem). Its protocols may simplify data transfers between CISM codes. The MxN problem refers to the difficulty in moving data from program A whose data is distributed over M processors to program B whose data is distributed over N processors. MetaChaos provides a prescription for determining the optimal method for accurately and efficiently moving the data between these applications by creating a linearization of the data distribution based upon the data descriptor provided to the MetaChaos library by the application developer. MetaChaos currently works with P++ data distributions. Dr. Sussman and his team are now working on infrastructure to support arbitrary data decomposition.

4) Validation and Metrics: H Spence led this discussion without the benefit of NCAR participation due to the erratic AG connection (and he will not be able to attend the next MI WG meeting on 4/18).

Model validation should be distinguished from metrics leading to a “skill score.” Harlan’s proposed that CISM models be validated against some common indices like Kp, AE, etc. as well as against selected direct measurements. Modelers should be involved in validating their own codes, but independent validation at BU will also be necessary. Data coordination for validation will be supported by BU though with somewhat limited resources. It is important to distinguish the different nature and quality of various types of data, e.g., model input data, model output data and data obtained from actual measurements.

Efforts should be made by CISM modelers to ensure that the output data streams from models are represented in a common format. M Wiltberger noted that if CISM modelers all use Open DX visualization software, it is relatively straightforward to specify a common output format from Open DX.

The question of which of the multitude of measurements should be used for validation was briefly considered. The availability of essentially continuous measurements from DMSP and ground-based magnetometers, and their relative ease of access, make these data particularly desirable. However, it was suggested that a ranking of the various measurement data streams be developed to determine their overall utility in validation studies. Various attributes of the data, e.g., quality of the measurement, ease of access, directly comparable with model inputs or outputs, etc. should be considered as weights in determining an objective rank order. The variables exchanged between codes in the MI Coupling flowchart suggests some key quantities for validation studies.

Agenda Items for the 4/18 meeting:

1) First results from the 1-way and 2-way coupled LFM-TING models (Wang, Wiltberger)

2) First results from the 1-way coupled LFM-RCM models (Lyon, Toffoletto)

3) What is needed to validate MI coupling? (all modelers)

4) Identification of metrics for assessing quality of coupled models (all modelers)

5) Send additional agenda items to william.lotko@dartmouth.edu

Submitted by W. Lotko, 3/17/03