Senior Consultant, Research & Development
Business Area Multiphysics
Fraunhofer Institute for Algorithms and Scientific Computing SCAI
Schloss Birlinghoven 1
53757 Sankt Augustin
Germany
Phone: +49 2241 14-4066
Previous Responsibilities at SCAI:
2008 - 2012 Group Leader Computational Chemical Engineering
2007 - 2010 Spokesman, Project Management Task Force
2006 - 2008 Senior Scientist
A profound understanding of the microscopic behavior of chemical systems is of substantial importance both for material and drug design. Computer simulations can contribute essentially to this purpose. We work on problems in the design and the investigation of polymer and protein systems and develop corresponding software and algorithms.
V.A. Harmandaris, D. Reith, N.F.A. van der Vegt und K. Kremer. Comparison between coarse-graining models for polymer systems: Two mapping schemes for polystyrene. Macrom. Chem. and Phys. 208: 2109, 2007.
T. Köddermann, D. Paschek und R. Ludwig. Molecular dynamics simulations of ionic liquids: A reliable description of structure, thermodynamics and dynamics. Chem. Phys. Chem., 8: 2464-2470, 2007.
A. Maaß, E.D. Tekin, A. Schüller, A. N. Palazoglu, D. Reith, and R. Faller. „ Folding and unfolding characteristics of short beta strand peptides under different environmental conditions and starting configurations”, BBA – Proteins and Proteomics 1804: 2003–15, 2010.
D. Reith and K. N. Kirschner, “A modern workflow for force-field development – bridging quantum mechanics and atomistic computational models.,” Comp. Phys. Comm. 182: 2184, 2011.
T. Köddermann, K. N. Kirschner, J. Vrabec, M. Hülsmann, and D. Reith, “Liquid–liquid equilibria of dipropylene glycol dimethyl ether and water by molecular dynamics,” Fluid Phase Equilibria 310: 25, 2011.
An appropriate modelling of the occurring force fields is a precondition for reliable molecular dynamics simulations. For this purpose SCAI develops algorithms and program packages (Wolf2Pack und GROW), which semi-automatically optimizes the force field paramters that have to be determined.
Generally molecular dynamics simulations require the adaptation of different atom or molecular models to desired physcial or chemical properties. In this procedure the appropriate choice of force fields and of corresponding parameters is a crucial issue.
SCAI develops modular program packages, which can be simply tuned to the different occurring optimization problems. Once started, the programs run mostly without user interaction until the force field ist generated. In this way, force field parameters can be determined for complicated atomistic and coarse grained models of any chemical matter.
Specifically, we created the following products:
Before molecular dynamics computer simulations can be executed, appropriate models have to be developed. This process consists of two fundamental steps: First, a general model of the occurring force fields is designed, then the free parameters of this model have to be determined. Within a model, different matters are distinguished only by their parameters. These parameters define thus the shape of the force field. The choice of these parameters is often more important for the quality of the chosen model than its analytical form.
It turns out that the parameter opimization for concrete systems is very time consuming. Therefore, often so-called standard force fields are used, which are known from the literature or contained in existing simulation software (e.g. GROMOS, CHARM or AMBER). However, there are many applications for which these common force fields are not well suited. Either the force fields are not optimized at all or only for certain matter classes (e.g. proteins) or adapted empirically under certain conditions (e.g. aqueous solution). The results of computations which have been based on such force fields may not be very reliable in applications with different conditions.
To use the program being developed by SCAI, the user decides on a suitable generic model. The corresponding free parameters are set to reasonable initial values. Using this first force field a simulation is run and analyzed comparing sensitive indicators (the target function) with available reference values. Thus an impression of the quality of the current force field is obtained and a relation to physical quantities is established. The algorithm optimizes now the current parameters automatically in further iterative simulation runs.
The automatic optimization of model parameters is going to become an indispensable tool of our research. In the atomistic level it allows for the fast adaptation of force fields, which are tailored for the system of interest, its surrounding conditions and the particular question of interest. The use of general »standard force fields« can thus be avoided.
In the field of coarse grained polymer models the model development is simplified significantly. We can concentrate more on the general model development and leave the optimization of each model to the computer. This accelerates, for example in the field of soft matter research, the development of new materials.
D. Reith, H. Meyer, and F. Müller-Plathe: ``Mapping Atomistic to Coarse-grained Polymer Models'', Macromolecules 34, 2335-45 (2001).
D. Reith, M. Pütz, and F. Müller-Plathe: ``Deriving effective mesoscale potentials from atomistic simulations'', J. Comp. Chem. 24, 1624-36 (2003).
V.A. Harmandaris, D. Reith, N.F.A. van der Vegt, and K. Kremer: “Comparison Between Coarse-Graining Models for Polymer Systems: Two Mapping Schemes for Polystyrene”, Macromol. Chem. Phys. 208, 2109 (2007).
M. Hülsmann, T. Köddermann, J. Vrabec, and D. Reith: “GROW: A Gradient-based optimisation workflow for the automated development of molecular models”, Computer Physics Communications 181, 499–513 (2010).
D. Reith and K. N. Kirschner, “A modern workflow for force-field development – bridging quantum mechanics and atomistic computational models.,” Comp. Phys. Comm. 182, 2184 (2011).
The computer simulation of polymer and protein systems has to face a scale problem, i.e. the interesting quantities require different scales of detail of the problem under consideration, reaching from the level of a single atom to the overall behavior of a polymer chain or even a polymer network. The time and length scales involved in a given problem can span several orders of magnitude. For the investigation of complex systems and effects (e.g. adsorption at surfaces, membranes or composites), it is necessary to simulate very large numbers of particles which means reaching fast the technically feasible due to limited computing resources. Here, the technique of Mesoscale Coarse Graining opens up new possibilities. The software package ESPResSo++ which we co-developed makes this technique available for efficient simulations.
By the method of Mesoscale Coarse Graining it is possible to significantly increase the simulated time and the size of chemical systems in computer simulations. This improvement results in more reliable predictions for physical quantities and properties which are indispensable for the development of new materials.
For this purpose, the atomistic, fully detailed system ist mapped to a coarser "mesoscopic" model. It is important to represent the omitted degrees of freedom in a suitably averaged way in the coarsened system such that the chemical identity of the substance under consideration is maintained. This can be guaranteed by reproducing key properties of the system.
Many macroscopic polymer properties can only be understood after a deep study of microscopic details. In general, computer simulations are an appropriate tool to turn microscopic basic knowledge in macroscopic information. For this purpose, quantum chemical computations or experimental data are used to construct a model which takes all involved atoms into account.
Unfortunately, in simulations based on this approach, problems may arise by the detailed treatment of the fast degrees of freedom (e.g. bond vibrations), which require that the time propagation of the system can proceed only in tiny little steps. The full simulated time is thus also very short (typically some nanoseconds) - even if high performance computers are used for weeks. This may even result in applications in which the slow degrees of freedom (e.g. the radius of gyration) do not reach the state of equilibrium at all which means that some properties remain unpredictable.
Very large atomistic systems cannot be simulated due to too large computing times either. Coarsened approaches ("coarse graining") offer one way of overcoming these dilemmas: with this method the fast degrees of freedom can be eliminated from the system, the time steps in the computation can be enlarged and the slow degrees of freedom can thus be relaxed. Using greater time and length scales are reached, which makes it possible to gain more information on a system and to find explanations for macroscopic physical effects. This is a decisive progress for the generation of new materials.
In the case of polyacrylic acid (PAA) in solution, data for of atomistic simulations were available for an oligomer. This system was coarsened in the following way: a full structural component, i.e.-CH2-CH1(-COOH)-ä was replaced by a single sphere around its center of mass (see. Fig. 1, where H-atoms are omitted). If this procedure is carried out for all structural components, a new coarsened chain is obtained (see Fig. 2). If it is possible, to reproduce the behavior of the atomistic chain, this model can be used to study longer PAA chains.
On the coarser level static effects such as adsorption at organic surfaces can be investigated much better because the free energy is influenced substantially by entropic contributions. Moreover, it is possible to re-introduce the atomistic details and to relax the fast degrees of freedom of the system if this is needed to understand certain effects.
A modern software package which makes the benefits of Mesoscale Coarse Graining available, is ESPResSo++.
D. Reith, H. Meyer, and F. Müller-Plathe: ``Mapping Atomistic to Coarse-grained Polymer Models'', Macromolecules 34, 2335-45 (2001).
D. Reith, B. Müller, F. Müller-Plathe, and S. Wiegand: ``How does the chain extension of poly (acrylic acid) scale in aqueous solution? A combined study with light scattering and computer simulation'', J. Chem. Phys. 116, 9100-6 (2002).
R. Faller and D. Reith: ``Properties of Poly (isoprene) - Model Building in the melt and in solution'', Macromolecules 36, 5406-5414 (2003).
V.A. Harmandaris, D. Reith, N.F.A. van der Vegt, and K. Kremer: “Comparison Between Coarse-Graining Models for Polymer Systems: Two Mapping Schemes for Polystyrene”, Macromol. Chem. Phys. 208, 2109 (2007).
Strategic planning based on reliable forecasts is an essential key ingredient for a successful business management within a market-oriented company. This is especially true for the automobile industry, as it is one of the most important sectors in many countries. Reliable forecasts cannot be fully based on intuitive economic guesses of the market development. Mathematical models are a natural key to increase the accuracy of the predictions as well as to guarantee for explicable, deterministic, and efficient calculations.
Methods based on time series analysis and statistical learning theory are powerful instruments to get insight into internal relationships within huge empirical datasets. They are able to produce reliable and even highly accurate forecasts on the automobile sales market. In addition, data mining algorithms have become more and more specific and complex in recent years. When utilizing them here, the accuracy of the prediction has the same importance as the explicability of the model.
In order to compose a meaningful model, a set of useful input parameters is needed. They include the registrations of new automobiles as well as economic exogenous parameters like the Gross Domestic Product, the Consumer Price Index, the Unemployment Rate, the Interest Rate, and the Industrial Investment Demand (alternatively considered on a yearly, quarterly, or monthly level).
Our self-developed current approach consists of a combination of time series analysis and data mining techniques. New car registrations are considered as a time series composed additively of trend, seasonal and calendar components, which are assumed to be independent. The seasonal and calendar components are estimated by standard time-series analysis methods, whereas the trend component is multivariate, i.e. it depends on economic exogenous parameters and is therefore estimated by data mining techniques like Support Vector Machines, Decision Trees, K-Nearest-Neighbor, or Random Forests. Fraunhofer SCAI now offers a software tool written in S, a scripting language related to the open-source R project for statistical computing. For the German Market, the forecast methodology delivers very low prediction error rates on a validation set (e.g. 4-6% for quarterly data).
However, reliable estimates for the exogenous parameters have to be provided by market experts in order to achieve accurate forecasts for the future. Also, the accuracy of the predictions is limited by the occurrence of special effects like financial crises or the car-scrap bonus, which also need to be incorporated meaningfully by market experts. Hence, this project is an interdisciplinary challenge between mathematical modeling and economy.
B. Brühl, M. Hülsmann, D. Borscheid, C.M. Friedrich, and D. Reith: “A Sales Forecast Model for the German Automobile Market Based on Time Series Analysis and Data Mining Methods.” In: Perner, Petra (Hrsg.): Advances in data mining : applications and theoretical aspects. pp. 146-160, Springer (2009).
M. Hülsmann, C.M. Friedrich, D. Borscheid, and D. Reith: “A Sales Forecast Model for Automobile Markets Based on Time Series Analysis and Data Mining Methods,” Advances in data mining: applications and theoretical aspects. (P. Perner, ed.), pp. 255-269, Springer (2011).
M. Hülsmann, D. Borscheid, C. M. Friedrich, and D. Reith, “General sales forecast models for automobile markets and their analysis,” International Journal Transaction on Machine Learning and Data Mining 5, 65-86 (2012).
1. F. Müller-Plathe and D. Reith: ``Cause and effect in reversed in non-equilibrium molecular dynamics: An easy route to transport coefficients'', Comput. Theor. Polym. Sci. 9, 203-209 (1999).
2. D. Reith and F. Müller-Plathe: ``On the Nature of Thermal Diffusion in Binary Lennard-Jones Liquids'', J. Chem. Phys. 112, 2436-43 (2000).
3. H. Meyer, O. Biermann, R. Faller, D. Reith, and F. Müller-Plathe: ``Coarse Graining of Nonbonded Interparticle Potentials Using Automatic Simplex Optimization to Fit Structural Properties'', J. Chem. Phys. 113, 6265-75 (2000).
4. D. Reith, H. Meyer, and F. Müller-Plathe: ``Mapping Atomistic to Coarse-grained Polymer Models'', Macromolecules 34, 2335-45 (2001). DOI: 10.1021/ma001499k
5. D. Reith, T.Huber, F. Müller-Plathe, and A. Torda: ``Free Energy Approximations in Simple Lattice Proteins'', J. Chem. Phys. 114, 4998-5005 (2001).
6. P. Bordat, D. Reith, and F. Müller-Plathe: ``The influence of interaction details on the thermal diffusion in binary Lennard-Jones liquids'', J. Chem. Phys. 115, 8978-8982 (2001).
7. D. Reith, B. Müller, F. Müller-Plathe, and S. Wiegand: ``How does the chain extension of poly (acrylic acid) scale in aqueous solution? A combined study with light scattering and computer simulation'', J. Chem. Phys. 116, 9100-6 (2002). DOI: 10.1063/1.1471901
8. B. Dünweg, D. Reith, M. Steinhauser, and K. Kremer: ``Corrections to Scaling in the Hydrodynamic Properties of Dilute Polymer Solutions'', J. Chem. Phys. 117, 914-24 (2002). DOI: 10.1063/1.1483296
9. D. Reith, H. Meyer, and F. Müller-Plathe: ``CG-OPT: A Software Package for Automatic Force Field Design'', Comput. Phys. Commun. 148, 299-313 (2002).
10. P. Bordat, J. Sacristan, D. Reith, S. Girard, A. Glättli, and F. Müller-Plathe: ``An Improved Dimethyl Sulfoxide Force Field for Molecular Dynamics Simulations'', Chem. Phys. Lett. 374, 201-5 (2003). DOI: 10.1016/S0009-2614(03)00550-5
11. R. Faller and D. Reith: ``Properties of Poly (isoprene) - Model Building in the melt and in solution'', Macromolecules 36, 5406-5414 (2003). DOI: 10.1021/ma025877s
12. D. Reith, M. Pütz, and F. Müller-Plathe: ``Deriving effective mesoscale potentials from atomistic simulations'', J. Comp. Chem. 24, 1624-36 (2003). DOI: 10.1002/jcc.10307
13. V.A. Harmandaris, D. Reith, N.F.A. van der Vegt, and K. Kremer: “Comparison Between Coarse-Graining Models for Polymer Systems: Two Mapping Schemes for Polystyrene”, Macromol. Chem. Phys. 208, 2109 (2007). DOI: 10.1002/macp.200700245
14. T. Müller, S. Roy, W. Zhao, A. Maaß, and D. Reith: ”Economic simplex optimization for broad range property prediction: Strengths and weaknesses of an automated approach for tailoring of parameters”, Fluid Phase Equilibria 274, 27-35 (2008). DOI: 10.1016/j.fluid.2008.06.009
15. B. Brühl, M. Hülsmann, D. Borscheid, C.M. Friedrich, and D. Reith: “A Sales Forecast Model for the German Automobile Market Based on Time Series Analysis and Data Mining Methods.” In: Perner, Petra (Hrsg.): Advances in data mining : applications and theoretical aspects. Berlin: Springer, 2009, pp. 146-160. DOI: 10.1007/978-3-642-03067-3%5F13
16. M. Hülsmann, T. Köddermann, J. Vrabec, and D. Reith: “GROW: A Gradient-based optimisation workflow for the automated development of molecular models”, Computer Physics Communications 181, 499–513 (2010). DOI:10.1016/j.cpc.2009.10.024
17. M. Hülsmann, J. Vrabec, A. Maaß, and D. Reith: “Assessment of Numerical Optimization Algorithms for the Development of New Molecular Models.” Computer Physics Communications 18, 887–905 (2010). DOI:10.1016/j.cpc.2010.01.001
18. A. Maaß, L. Nikitina, T. Clees, K.N. Kirschner, and D. Reith: “Multi-objective force field optimization on basis of random models for ethylene oxide” Molecular Simulation 36, 1208-18 (2010). DOI: 10.1080/08927020903483312
19. K.N. Kirschner, K. Heikamp, and D. Reith: “Atomistic Simulations of Isotactic Poly(Methyl Methacrylate) Melts: Exploring the Backbone Conformation.” Molecular Simulatios 36, 1253-64 (2010). DOI: 10.1080/08927020903536374
20. M. Hülsmann, T. J. Müller, T. Köddermann, and D. Reith: „Automated Force Field Optimisation of Small Molecules using a Gradient–Based Workflow Package.” Molecular Simulation 36, 1182-96 (2010). DOI: 10.1080/08927022.2010.513974
21. A. Maaß, E.D. Tekin, A. Schüller, A. N. Palazoglu, D. Reith, and R. Faller. „ Folding and unfolding characteristics of short beta strand peptides under different environmental conditions and starting configurations”, BBA – Proteins and Proteomics 1804, 2003–15 (2010). DOI:10.1016/j.bbapap.2010.06.019
22. D. Reith and K. N. Kirschner, “A modern workflow for force-field development – bridging quantum mechanics and atomistic computational models.,” Comp. Phys. Comm. 182, 2184 (2011). DOI:10.1016/j.cpc.2011.05.018
23. T. Köddermann, K. N. Kirschner, J. Vrabec, M. Hülsmann, and D. Reith, “Liquid–liquid equilibria of dipropylene glycol dimethyl ether and water by molecular dynamics,” Fluid Phase Equilibria 310, 25 (2011). DOI: 10.1016/j.fluid.2011.07.015
24. M. Hülsmann, C.M. Friedrich, D. Borscheid, and D. Reith: “A Sales Forecast Model for Automobile Markets Based on Time Series Analysis and Data Mining Methods,” Advances in data mining: applications and theoretical aspects. (P. Perner, ed.), pp. 255-269, Springer (2011).
25. M. Hülsmann, D. Borscheid, C. M. Friedrich, and D. Reith, “General sales forecast models for automobile markets and their analysis,” International Journal Transaction on Machine Learning and Data Mining 5, 65-86 (2012). ISBN: 978-3-942952-12-5
26. A. Wolf, D. Reith, and K. N. Kirschner, "Thiopeptide antibiotics and the ribosomal 23s-l11 subunit - a challenging use case for semi-automatic force-fi eld development," in Proceedings for the Computational Biophysics to Systems Biology 2011 Workshop in Jülich (P. C. et al., ed.), vol. 8, pp. 65-69, IAS Series, Forschungzentrum Jülich GmbH, 2012
27. T. Brandes, A. Arnold, T. Soddemann, and D. Reith, “CPU vs. GPU - Performance comparison for the Gram-Schmidt method," Eur. Phys. J. ST 210, 73-88 (2012). DOI: 10.1140/epjst/e2012-01638-7
28. J. Halverson, T. Brandes, O. Lenz, A. Arnold, S. Bevc, K. Kremer, T. Stühn, and D. Reith, "ESPResSo++: A modern multiscale simulation package for soft matter systems," Comp. Phys. Comm. 184, 1129–1149 (2013). DOI: 10.1016/j.cpc.2012.12.004
29. M. Hülsmann, S. Kopp, M. Huber, and D. Reith, "Efficient gradient and hessian computations for numerical optimization algorithms applied to force fi eld developments with molecular simulations." J. Phys.: Conf. Ser. 410, 012007 (2013). DOI:10.1088/1742-6596/410/1/012007
30. M. Hülsmann, S. Kopp, M. Huber, and D. Reith, "Utilization of efficient gradient and hessian computations in the force field optimization process of molecular simulations." Comput. Sci. Disc. 6, 015005 (2013). DOI:10.1088/1749-4699/6/1/015005
31. O. Krämer-Fuhrmann, J. Neisius, N. Gehlen, D. Reith, and K. Kirschner, "Wolf2pack - Portal based atomistic force-fi eld development," J. Chem. Inf. Mod. 53, 802–808 (2013). DOI: 10.1021/ci300290g
32. T. Köddermann, D. Reith, and R. Ludwig, "Force Field comparison on various model approaches – how to design the best model for the ionic liquid family [CnMIM][NTf2]," Chem. Phys. Chem. 14, 3368-3374 (2013). DOI: 10.1002/cphc.201300486
33. T. Köddermann, D. Reith, and A. Arnold, "Calculating concentration-dependent Log Pow values from Atomistic Computer Simulations," J. Chem. Phys. B ,117, 10711-10718 (2013). DOI: 10.1021/jp405383f
34. M. Hülsmann and D. Reith, "SpaGrOW – A derivative-free optimization scheme for intermolecular force field parameters based on sparse grids methods," Entropy 15, 3640-3687 (2013). DOI: 10.3390/e15093640
35. A.A. Tietze, F. Bordusa, R. Giernoth, D. Imhof, T. Lenzer, A. Maaß, C. Mrestani-Klaus, I. Neundorf, K. Oum, D. Reith, A. Stark, "On the nature of interactions between ionic liquids and small amino acid-based biomolecules", Chem.Phys.Chem. (2013), im Druck.
B. Other Science-related Publications
1. D.Reith: "The Uppsala ESI TOF mass spectrometer", ISSN 0284 -2769, TSL/ISV-96-0159, Uppsala University Press (1996).
2. D.Reith: "Thermal Diffusion in binary Lennard-Jones liquids" , Diploma thesis, University of Mainz (1998).
3. D. Reith: "Die Seele des Kunststoffs erforschen", Invited Contribution in „Mainzer Allgemeine Zeitung“, Saturday, Feb 19 2000.
4. O. Biermann, R. Faller, H. Meyer, F. Müller-Plathe, D. Reith, and H. Schmitz: ``Automatisiertes Erstellen von Simulationsmodellen für Flüssigkeiten und Polymersysteme'', , in: ``Forschung und wissenschaftliches Rechnen -- Beiträge zum Heinz-Billing-Preis'', Th. Plesser and H. Hayd (eds.), GWDG, Göttingen, Bericht Nr.56, 49-63, ISSN 0176-2516 (2001).
5. D.Reith: "Neue Methoden zur Computersimulation von Polymersystemen auf verschiedenen Längenskalen und ihre Anwendungen" , PhD thesis, University of Mainz (2001).
6. D. Reith: "Von der Fastnacht über Lachs zum Stern - Beispiel eines naturwissenschaftlichen Studiums ", Invited Contribution, Cronicles of Bischöfliches Cusanus-Gymnasium Koblenz, Academic Year 2001/02.
7. M. Hülsmann, C.M. Friedrich, D. Reith: „A New Approach to Explicable Sales Forecast Models for the German Automobile Market.” ERCIM News, 78: 30-31 (2009).
8. M. Hülsmann, T. Köddermann, D. Reith: „Engineering chemical substances via molecular simulations utilizing efficient gradient-based optimization algorithms.“, ERCIM News, 81: 25-26 (2010).
9. D. Reith and K. N. Kirschner: "Mathematics meets Chemistry - Workflow-guided evolving software for Molecular Modelling.", ERCIM News, 88: 47-48 (2012).
Fraunhofer Institute for Algorithms and Scientific Computing SCAI