SBE&S Vision

Dear Folks:

(I realize that this construction reveals me to be hopelessly out of date, and inconsistent with this cutting edge communication medium, but very old habits will go with me to the grave.)

The commenting system seems to me to be what we need. It worked pretty smoothly for me. The only glitch is that after reading the Terms of Service, I was bumped out of the system before I could click on Accept and complete my registration. It went through anyway, maybe that's a bad thing.

Here's a couple of substantive suggestions:

On the WTEC reports page the only choice for the SBES report is the B&W version. I think the high-res color version would be better.

I don't see how to view the other comments on the Welcome to SBE&S vision page. Maybe that's because there ain't any yet.

duane

I wonder if System Dynamics (which is different from Systems Thinking but related to it) can be included in the vision of the future of SBE&S. The methodology of System Dynamics is quite developed, the software is already available, many applications have been developed, the books have been published and the syllabi are ready. And the best part about System Dynamics is that the modeling itself is very user-friendly — there is no need to learn C++, Java, etc. Here are a couple of links:

http://www.systemdynamics.org/
michaeljosephradz…om/sdlinks.htm

Best regards,
Oleg

Oleg,

As a quick reply to your post, the Innovation and Engineering Design group explicitly included Systems Dynamics in its list of recommendations. I'd encourage you to help us develop the concept.

Best regards,

Loren

The report resonates strongly with what I have been thinking about the impact of high performance computation on materials research. There is great opportunity, but it is clear that in current funding situation, these opportunities will not be realized in the US. Several points, perhaps not brought out in the report:

I think the emphasis could be stronger on new electronic structure methods. This is both a key intellectual and practical bottleneck: namely, how to use HPC to calculate properties of correlated quantum systems to sufficient accuracy. The discussion centered on either semi-empirical or DFT based methods, which is natural sinsc that is what has made an impact to date. However, both have serious deficiencies, e.g. the discussion of quantification of errors on pages 91,97-98. Methods to directly compute forces from the basic equations are around, such a coupled cluster and quantum Monte Carlo methods. In the future, I think there will be tremendous challenges in understanding electronic properties of materials, particularly those involving transition metals, e.g. magnetic, optical and transport properties. I think simulations can make a great contribution in bringing new materials, designed to have particular properties into applications. However, the algorithms, such as DMFT, QMC, GW and codes need much work. Note that it took 30 years for DFT methods to get to the present state. The planning needs to have a longer time scale in mind, since these are difficult but very important problems with much economic payback.

The second point is the management of software development in the US. It is clear that it has been ineffective in the last few decades. Though there has been funding, that has not led to the US capitalizing on the invention of the algorithms or the substantial funding in computer hardware. Many of the problems have been pointed out in the report. The reports on the successful European efforts are very valuable in this respect. We need to plan how to change the US model for software development. I think some of the ingredients will be:

Dear Loren,

I'd love to learn more about your project and about how I can contribute. Whom should I contact?

Thanks,
Oleg

I am going to chime in here?probably as the token liberal arts graduate. But I?d like to comment on a broader policy opportunity.

It struck me as a bit odd that we are putting together a road map for SBE&S and there were only two industrial representatives at this meeting. And yet we know from many studies by the Council on Competitiveness that for the companies that have embraced modeling and simulation, it is essential to their business survivial?they simply can?t compete in the market without it.

We also know that this is not penetrating the supply chains of many of these companies . Unless this is resolved, our more experienced industrial users in manufacturing will not be able to do full system simulations as they will lack important models from their suppliers…a true tail wagging the dog situation. And the suppliers will fall behind any competition that adopt SBE&S. Bad news for the economy and national security.

I would to suggest a national campaign to make modeling and simulation with high performance computing a ?best business practice? ?whether that business is in the commercial world or the world of university or laboratory research. I think there is an analogy, at least in the business world, that applies. And that is the quality movement.

In the late 1980?s-early 1990?s, corporate America had not embraced ?quality? like Japan had and we were losing global market share in many industries as a result. With the leadership of a few companies (Motorola for example), a national quality movement began, and the government participted…with its ?bully pulpit? (Commerce created the Malcolm Baldridge Quality Award for example.)and also its own quality movement. It took time?but with persistence and leadership eventually there was widespread buy in, universities began teaching classes in six sigma etc. Now it is embedded in the way we do business.

I?m concerned that unless there is a similar coordinated ?campaign? approach, SBE&S will not get widespread adoption…or not in the timeframe needed to maintain our competitiveness.

I'd like to add that the commercial ISVs must be considered in any national software dialog and included somehow. Many companies rely on ISV code and are not able to internalize university or lab codes into their production environments. ISVs are struggling yet are an important part of the industrial HPC community. You might look at the Council report "Accelerating Innovation for Competitive Advantage: The Need for Better HPC Application Software Solutions" It's a bit dated but many of the ISV issues discussed are still current. http://www.compete.org/p…3/accelerating-innovation-for-competitive-advantage-the-need-for-better-hpc-application-software-solutions/

In the not too distant future, we wish to a design a software system for the discovery of new enabling materials. For discussion purposes, we focus on the discovery of new materials for the design of an engineering structure that can perform its prescribed functions under harsh environmental conditions within a predetermined lifetime. A group of scientists and engineers starts with a prototype materials sample and uses advanced experimental and imaging techniques to create digital data sets of material microstructure evolutions. Using the to-be-developed ?Multiresolution Data Sets Mathematical Theory?, the ?predictive science based governing laws of the materials microstructure evolutions? are derived and melted into a ?Stochastic Multiresolution (multiscale) Design Framework.? We predict that the Future software system will be constructed based on a Probabilistic Computational Science-based Mathematical Framework of which its verification and validation are done through limited experiments, and the LIFE-CYCLED MATERIALS DISCOVERY AND DESIGN for products design and manufacturing is performed through the use of high performance computing.

Man made functional materials often do not have defined governing laws for their mechanical, chemical, biological, electrical and thermal properties. By combining mathematics and physical sciences, computer science, image processing, engineering, and experimental results, we want to derive the governing laws for the materials microstructure evolutions via a multiresolution data set analysis. Experimental techniques can create digital data sets representative of the microstructure. These data sets extend down to the quantum mechanical scale and can often be generated from first principles calculations. We plan to map these data sets of microstructures to a continuous resolution axis to make the varying materials length scales more accessible. The life-cycle microstructure evolution is subjected to harsh environments (radiation, thermal-mechanical, etc). We want to recast this microstructure evolution as a mathematical stochastic design and discovery problem. Since these problems can potentially become very complex, the solutions may require clever mathematics, materials science, and new experiments to build a solid foundation.

Our vision for deriving the governing laws of life cycle functional materials discovery and design is to treat large experimental and computational data sets as the building blocks. The derivation of the mathematical model will involve testing, imaging, characterization, verification and validation, and uncertainty qualification. While physical science and experimental observation will heavily influence the model, it will ultimately be the new predictive science mathematical theory that will unify these ideas and lead to a solution.

We envision that the data set mathematical foundation will start with linking spatial scales for continuous resolution of a microstructure. We want to be able to zoom into a microstructure in the same way that modern satellite technology allows us to zoom into images anywhere, anytime, and with any resolution. Hence, the separation of data by scales is done through the use of computer imaging and materials science knowledge. The quantification and variability of microstructured data sets are performed through the use of statistical and decision making theories, whereas the scale linking is done via testing and characterization of the data set samples. We then have refined data in order of scale, and these data sets are mapped into the resolution axis, which is an extension of the length scale theory in mechanics of materials, a variable length scale theory with gradients. Note that the data is in 4 dimensions, (x,y,z,t), so the resolution axis is the 5th dimension. Now the microstructure is shown in continuous resolution. To represent the stochastic nature of the data, additional dimensions will be added for each parameter with axes representing their distributions.

Finally, to establish the governing laws of these very heterogeneous microstructure evolutions subject to extreme environments, we start with the known single scale homogeneous thermal-mechanical, electrical and mass diffusion laws, etc. We then need to use multi-scale physics to design better multi-functional materials and use the above refined multiresolution data sets to develop a technique to extract the missing information that would otherwise remain hidden in the results of the carefully designed experiments. A good start is to link the missing science with existing single scale governing laws by introducing microstructure fluxes, identifying microstructure transition events, and linking scales by introducing microstructure couple fluxes.

References
Liu WK, Karpov EG, Zhang S, Park HS. An Introduction to Computational Nano Mechanics and Materials, Computer Method in Applied Mechanics and Engineering 193(17-20), 1529-1578, 2004.

Wing Kam Liu, Yaling Liu, David Farrell, Lucy Zhang, X. Sheldon Wang, Yoshio Fukui, Neelesh Patankar, Yongjie Zhang, Chandrajit Bajaj, Junghoon Lee, Juhee Hong, Xinyu Chen, and Huayi Hsu, ?Immersed Finite Element Method and Applications to Biological Systems?, Computer Methods in Applied Mechanics and Engineering, 195(1722-1749), 2006, (One of the ten most cited articles 2005-2008 published in Computer Methods in Applied Mechanics and Engineering).

Albert C. To, Wing Kam Liu, Gregory B. Olson, Ted Belytschko, Wei Chen, Mark S. Shephard, Yip-Wah Chung, Roger Ghanem, Peter W. Voorhees, David N. Seidman, Chris Wolverton, J. S. Chen, Brian Moran, Arthur J. Freeman, Rong Tian, Xiaojuan Luo, Eric Lautenschlager, A. Dorian Challoner, ?Materials integrity in microsystems: a framework for a petascale predictive-science-based multiscale modeling and simulation system,? Computational Mechanics, Volume 42, Number 4, September, 2008, pp. 485-510.

Wing Kam Liu, Larbi Siad, Rong Tian, Sanghoon Lee, Dockjin Lee, Xiaolei Yin, Wei Chen, Stephanie Chan, Gregory B. Olson, Lars-Erik Lindgen, Mark F. Horstemeyer, Yoon-Suk Chang, Jae-Boong Choi and Young Jin Kim, ?Complexity science of multiscale materials via stochastic computations,? DOI: 10.1002/nme.2578, 2009.

Observations and Comments

1. Paucity of industrial stakeholders at workshop and as panelists/advisors

Input from industry may not be significantly different from that of academia or the national labs, but, ultimately, it is different enough to affect the success of computational modeling in an industrial setting.

2. Misplaced emphasis on computer speed and supercomputers

In my 3 decades of computational work, computers have always gotten faster, and I expect even faster computers in the future. Computational researchers will always be in need of faster computers, more memory, and more storage space! Although never fully satisfied, they have managed admirably with their existing resources! So, I have lived through kilo-, mega-, giga-, and tera-scale computing and probably will be around to experience peta-, exa-, zetta-, and yotta-scale computing over the next few decades.

3. Key funding priorities:

a. Software development, implementation, interoperability, and maintenance

b. Education and training, which includes internship and scientist-exchange programs to enhance cross-fertilization between industry, academia, and national labs.

Let's be clear and honest … in the immediate future (10-20 years), software and people, not hardware, are the key components for successfully integrating computational modeling with both experimental and engineering technologies.

1. Creating Software for Creating Models: Languages, performance analysis and debugging
This is particularly important in areas where modeling is not mainstream, like in biomedical research.
2. Using the Model-Data Interface: Modeling paradigms, domains, simulations that require and/or generate data, especially large data sets.
Software for parameter identification in complex (non-linear, multi-variable) systems is particularly needed in biomedical research.
3. Workforce Development ? Education and training; broadening participation; virtual communities, gateways
Education and training in biomedical research is strongly needed, since this kind of training is not usually considered in biomedical areas.
4. Developing, Implementing and Extracting Knowledge from Models: New physical models and algorithms. Multiscale methods, visualization, validation, verification, uncertainty quantification.
These tools are particularly useful and needed in biomedical research, given the complexity of the biomedical systems and their difficulties in modeling, parameter identification and analysis.
5. Discovery by Simulation and Modeling: Success, future prospects; funding mechanisms to support SBE&S-based discovery
Predictive models in biomedical research, and their use in establishing hypotheses will be very useful. Agencies should explore ways to support hypotheses driven research that is based or highly supported by modeling and simulation (particularly modeling and simulation that is based on biomedical data).
6. Innovation and Engineering Design: SBE&S for optimization, design, multi-scale time-critical adaptive optimization like supply chain management and optimization
This can be a future task in biomedical modeling and simulation, since at present the mere task of getting modeling and simulation into biomedical research seems difficult enough.

1. Including atomic-scale physics and chemistry in engineering simulations is indeed a critical area for multiscale simulations. Similarly, overcoming the time- and size-bottleneck in biological simulations, i.e., combining quantum-mechanical, molecular mechanics and larger-scale models is also a critical area for SBE&S in the biomedical area.

2. The dominance of major software packages developed in Europe, rather than in the U.S., is likely due to our funding structure at universities, which uses 3-year funding cycles, These do not encourage long-term software development and maintenance activity. While national labs have a better record in long-term funding of software development, their efforts in unclassified areas have been mainly focused on a system software and software tools, rather than application packages that would benefit individual science or engineering areas.

3. Since software development activities are long-term efforts, the core groups must include long-term and permanent employees. At present, this structure challenging to establish at universities, where individual professors lead groups of graduate students and postdocs. Perhaps longer-term center-level grants and joint projects between universities and national labs would help.

4. A correction: the development and maintenance of Quantum Espresso is headquartered at SISSA (Trieste, Italy) and headed by Stefano Baroni. The original PWCSF and Car-Parrinello codes were also developed there.

The below is some text that I have slightly modified from material I prepared for an NAS committee examining promising approaches for broadening participation of underrepresented minorities. This may be more statistics/demographics heavy, and more focused on astro, than you might like, but I suspect you can easily repurpose it for the SBES context.

Workforce Development and Broadening Participation

Underrepresented Minorities in Data-intensive Astrophysics

Black, Hispanic, and Native Americans (underrepresented minorities; URMs) comprise 27% of the US population but are less than 4% of the astrophysics workforce. These PhD-granting programs today collectively award approximately 51 URM PhDs per year, an average per PhD-granting institution of 1 URM PhD every 10 years. This represents a modest increase in absolute number from 31 URM PhDs in 1988. The corresponding fraction of URM PhDs has been roughly flat at 2-4% of the total, while the proportion of URMs in the U.S. population grew by 33% during this same time period (from 20.9% in 1988 to 27.0% in 2008; data from US Census).

Engaging URM individuals from a broader base of science and engineering (STEM) backgrounds could substantially, and quickly, expand the pool of qualified individuals in areas of astronomy and astrophysics that are likely to experience growth in the coming decade. For example, the increasing importance of high-performance computing and informatics-based approaches — for large scale simulations, for data-intensive surveys, for data-mining infrastructures — will require expertise that might be tapped from the ranks of computer science and engineering graduates.

In 2006, for example, URMs earned a total of 17,813 baccalaureate degrees in physics, computer science, and engineering. In comparison, 3,598 (20.2%) of these earned a master's degree, and 292 (1.6%) went on to earn a PhD. Thus the pool of URMs with relevant STEM training is substantial, but an overwhelming majority of these individuals currently exit the higher education pipeline with a bachelor?s degree. The opportunity for recruitment of URM STEM baccalaureates into advanced degree programs in astronomy and astrophysics is large.

Large-scale efforts to broaden participation of URMs from physics and other STEM disciplines will likely require strategic educational and research partnerships involving minority-serving institutions (MSIs), as these institutions are highly productive sources of URM talent in STEM disciplines. For example, the top 15 producers of African American physics baccalaureates in the US are all HBCUs, and just 20 HBCUs were responsible for producing fully 55% of all African American physics baccalaureates in the US between 1998 and 20075. In comparison to majority institutions, which in 2006 produced on average 9.0 URM bachelor?s degrees per institution per year in physics, computer science, and engineering, MSIs produced on average 36.1 URM degrees per institution per year in these disciplines. Moreover, these institutions are successful at placing students in PhD programs. Among the US baccalaureate-origin institutions of African American STEM PhD recipients, the top 8, and 20 of the top 50, were HBCUs (for the years 1997-2006).

Importantly, the number of MSIs with research-active faculty, and that offer advanced STEM degrees, has undergone dramatic growth. The growth of MSI Master's degree programs in particular is striking. For example, between 1987 and 2006, the number of MSIs offering Master's degrees in physics, computer science, and engineering increased by 79%, and the number of URMs earning Master's degrees from these institutions increased correspondingly by 533% (from 119 URM degrees in 1987 to 753 in 2006). Consequently, URMs who earn PhDs in STEM fields are ~50% more likely than their non-minority counterparts to have earned a "terminal" master's degree (i.e. not a master's degree earned as part of a PhD program) before eventually transitioning to a PhD program7. Thus the Master?s degree is a critical, and previously poorly understood, transition point for many URMs.

These programs especially offer significant opportunities for the development of partnerships in which the MSI partners are equal stakeholders in mission-critical research and funding. At the same time, the added institutional transitions experienced by many URMs also represent additional "leaks" in the pipeline prior to the PhD. There is need for the development of explicit institutional links and bridges to attend to transitions in the career development pipeline, to further ensure that potential PhDs are not lost. Focused and sustained support is needed as individuals traverse these transition points, to ensure that they experience continuity of mentoring, support, and research engagement.

Promising examples have recently emerged, such as the Fisk-Vanderbilt Masters-to-PhD Bridge program through which students use the Master's degree at Fisk as a stepping stone to the PhD at Vanderbilt through collaborative research projects at the two institutions and at national centers. The Fisk-Vanderbilt Bridge program as of 2008 includes 30 URM graduate students with undergraduate training in physics, engineering, materials science, and computer science performing research in traditional astronomy areas as well as in astro-informatics through the Vanderbilt Initiative in Data-intensive Astrophysics (VIDA; http://www.vanderbilt.edu/astro/vida).

The symbol that is not appearing printing correctly in the above is +/-.

I define physics-based computational engineering as the use of computational applications that use the laws of physics to design physical objects, such as an airplane.
The first major barrier to the adoption of physics-based computational engineering application software is the time and money it takes to develop such software. Case studies indicate that it takes a large team of 15 to 30 professionals as long as ten years to develop a multi-physics design code.
The second major barrier is the ability to generate geometries and mesh them for analysis. This can take weeks to months to a year or more depending on the complexity of the object being designed.
The third major barrier is the challenge of developing scalable algorithms for massively parallel computers. Few engineering applications in use today can use 1000 processors, let alone the 10,000 to 100,000 processors available on today?s biggest computers. In ten years, workstations will have this capability, but we won?t be able to use it unless we do the research to develop the algorithms.
There are other barriers: the ability to analyze and visualize large data sets, the lack of sound software engineering practices specific to computational engineering, and the lack of good collaborative infrastructures that can support a distributed development environment.

I recommend that the federal government, particularly the basic research funding organizations like NSF, ARO, DARPA, DOE, NASA, AFOSR, ONR, NIST, etc., support the following research topics to enhance the US capability in computational engineering.

The recommendations include research and demonstration projects in the following key technical areas:
–Development of computational mathematics algorithms for solving equations on large-scale massively parallel computers
–Development of improved methods for the generation of geometries and meshes
–Development of improved methods and capability for data analysis and visualization in a distributed environment
–Development of improved methods for verification and validation
–Development of software engineering methods for the development of physics-based computational engineering tools
–Development of collaboration methods and tools for enhancing code development by distributed, non-collocated teams, especially in the present evermore restrictive computer security environment.

Your last comment was validated in research by the Council on Competitiveness….their surveys and reports found that that the #1 reason companies don't do more modeling and simulation with HPC is the lack of talent…people within their firms or accessible to them that understand how to do this. The #2 barrier is the lack of production quality application software.

Building on Kerwin Dobbs comment above that "Input from industry may not be significantly different from that of academia or the national labs, but, ultimately, it is different enough to affect the success of computational modeling in an industrial setting"…I suggest that if the agencies Doug mentions do in fact aree to fund proposals some of Doug's recommended areas, that they require industry participation in at least some of them in order to receive fuding…to be sure that industry's issues are addressed. Industry could be industrial users and/or ISVs.

Another idea would be to do the same thing in the SciDAC program

I work extensively with the Modelica Association. The goal of the Modelica Association is to advocate a standard non-proprietary modeling language for describing and exchange behavior of physical systems. In this context, "physical systems" is currently limited to lumped systems (DAEs and ODEs) but there is some interest to expand this into solving PDEs.

I see several frustrating problems around these issues. There is some really amazing work being done in this area demonstrating a deep grasp of not only the engineering involved but also the need for standards, interoperability and process. The problem is that almost all that work is being done in Europe. Most of the people I interact with in the US (both in industry and academia) are pretty ignorant of the methods involved and generally lack the deep understanding of the physical phenomena as well. I find this very troubling for the future of the US since it has made its mark by leading in innovation.

I want to echo the sentiment that faster computers are not the real solution here. Many computational limitations (at least in the areas I work on) can be solved by finesse (better methods) rather than pure CPU cycles. But beyond that, people need to be better educated in building models, testing models and integrating models into the engineering process.

From my perspective we need more work in the following areas:
* A much better model-based science, engineering and medical curriculum.
* More funding at the industrial level. Frankly, as an industry person I get very frustrating seeing large amounts of money going almost exclusively into academia with very little to show for it in terms of solving real problems. Now I also think that any such funding needs to be linked to a strategic agenda (see more below).
* Home grown competency in developing advanced algorithms. In my particular area this means things like DAE index reduction and mixed symbolic-numeric methods.
* Development of standards around which to build a curriculum, tools and technology. My interest in Modelica is because it has the rarely found dual benefit of being very useful for solving real world problems but also has a solid and non-proprietary foundation for building new algorithms and exploring new ideas.

I want to see the US continue as an engine for global innovation but I'm troubled by two things. First, I am very impressed by lots of work I see coming out of Europe and I don't see the US responding (effectively) to this. Second, I see industry leading academia in terms of innovation (e.g. development of algorithms) in my area which is also a troubling sign as well.

Sailor Sam…

http://288547.com/Skunkworks/index.php/member/37/ ok…

echelon ladder…

http://www.spiritatwork.org/index.php/forums/member/695/ ok…

Frederick…

http://www.fricked.com/index.php/member/41/ ok…

Eric…

http://subliminal.lunanina.com/member/496/ ok…

Peter…

http://healthcarebeyondthepolitics.org/index.php/member/182/ ok…

Jurgen…

http://www.dbnwo.pun.pl/profile.php?id=177 for sure…

Peter Inocule…

http://www.whsecoed.ca/member/19/ is a truth…

Frederick Right…

http://www.beanblag.com/index.php?/member/59/ is the way…

Wrong is right…

http://www.happyisyellow.com/index.php/member/15/ is hot…

Bongo Dongo…

http://www.pv.pun.pl/profile.php?id=81 echo…

Excitement27…

http://www.w4tweaks.pun.pl/profile.php?id=98 I think…

florist18…

http://www.smilepolitely.com/member/2517/ inside…

trudgewater…

http://exiled.dutchcelt.nl/member/37/ edgy…