A New Runge-Kutta-Fehlberg Burn-up Solver Charles A. Wemple 2011 International Users Group Meeting Stockholm, Sweden | May 24-25, 2011 Overview • Introduction • Overview of Burn-up Methods • Runge-Kutta-Fehlberg Method • Implementation of R-K-F Method • Test Results • Summary, Conclusions, and Future Work May 24-25, 2011 International Users Group Meeting Introduction • Lattice codes have two fundamental functions – Solve transport equation – Calculate isotopic transmutation (burn-up) • Great improvements in solving transport equation • Few improvements in solving burn-up equation – why? – Present methods work pretty well – Existing solver is quite fast – small fraction of total CPU time – More rigorous methods require more CPU time • Modern computers limit penalties for higher accuracy • Conclusion – time to look at burn-up solvers! May 24-25, 2011 International Users Group Meeting Burn-up – What are we solving? May 24-25, 2011 International Users Group Meeting Burn-up – What are we solving? (cont.) • Make approximation of “interval invariance” – – – – Divide time variable into multiple intervals Assume flux constant over each interval Then, reaction rates are also constant Only isotopic concentrations vary during interval • Reduces integro-differential to ordinary differential – Simpler to solve – With predictor-corrector, errors are essentially negligible May 24-25, 2011 International Users Group Meeting Burn-up Methods – A Trade-off Study • Chain Linearization – Extremely fast – Limited accuracy • Chain approximations • Limited reaction paths May 24-25, 2011 International Users Group Meeting Burn-up Methods – A Trade-off Study (cont.) • Matrix Exponential Method – – – – Widely used (ORIGEN, etc.) Well tested – many years of experience Matrix rapidly gets large – NxN, where N is number of nuclides included Numerical methods all have limitations - see Moler and VanLoan, SIAM Review, 45 (1), 3-49 • Runge-Kutta methods – – – – – Widely used for solving differential equations Great variety of solutions available Excellent numerical properties Very accurate (if applied properly) But…..fairly slow May 24-25, 2011 International Users Group Meeting Runge-Kutta-Fehlberg Method May 24-25, 2011 International Users Group Meeting Runge-Kutta-Fehlberg Method (cont.) May 24-25, 2011 International Users Group Meeting Runge-Kutta-Fehlberg Method (cont.) • For Runge-Kutta-Fehlberg, solve R-K twice – First for qth-order equations – Again for (q+1)th-order equations – Adjust step size, h, to force difference (local error) below defined level • Chose 4th-order R-K-F method • Selected widely-used coefficient set i ai bi0 bi1 bi2 bi3 bi4 0 1 1/4 1/4 2 3/8 3/32 9/32 3 12/13 1932/2197 -7200/2197 7296/2197 4 1 439/216 -8 3680/513 -845/4104 5 1/2 -8/27 2 -3544/2565 1859/4104 May 24-25, 2011 International Users Group Meeting -11/40 ci di 25/216 16/135 0 0 1408/2565 6656/12825 2197/4104 28561/56430 -1/5 -9/50 2/55 Implementation of R-K-F Method May 24-25, 2011 International Users Group Meeting Implementation of R-K-F Method (cont.) • Problem – how to define step size, h? – Iteration in standard R-K-F method is computationally inefficient – Testing showed solution stability dependent on just a few isotopes • • 105Ru and 105Rh for equilibrium I/Xe 135I and 135Xe for non-equilibrium – Step sizes longer than ~5x half-life produced instabilities • Initial solution – fixed step size – Equilibrium – h=75 000 sec – Non-equilibrium – h=25 000 sec – Decay – 32 steps per interval May 24-25, 2011 International Users Group Meeting Implementation of R-K-F Method (cont.) • Problem – still pretty slow – Inherently dependent on length of burn-up interval – Acceptable for short intervals; unacceptable for longer intervals • Testing demonstrated two important characteristics – Short steps necessary for start of burn-up interval – Longer steps after concentrations saturate • Solution – sub-stepping – – – – Break each burn-up interval into sub-intervals Very short steps for early sub-intervals; increase as interval progresses Isotope-dependent sub-interval and step definitions Sub-intervals and step sizes defined to reproduce fine-step results • Reduced computation time by ~3x May 24-25, 2011 International Users Group Meeting Test Results • R-K-F method implemented in HELIOS2 • Basic testing - variety of pin-cell calculations – – – – UO2 MOX UO2-Gd2O3 U-Th • Summary of results – Eigenvalue difference up to ~600pcm – Small differences in major isotope concentration - ~0.5% – Large differences in minor isotope concentrations May 24-25, 2011 International Users Group Meeting Test Results (cont.) • Yankee Rowe isotopics benchmark – 18x18 PWR with control blades – Westinghouse performed destructive pin analyses (1966 & 1969) – Isotopics for actinides only – U → Cm • Summary of results – – – – Eigenvalue differences less than 200pcm Small differences in main actinides – up to ~0.3% Larger differences in lesser actinides – up to several percent Improved agreement with measured concentrations May 24-25, 2011 International Users Group Meeting Summary, Conclusions, and Future Work • Developed new Runge-Kutta-Fehlberg burn-up solver • Test implementation in HELIOS2 • Improved isotopic concentration results • Slower, but more accurate, than linearized chains • Will be released in HELIOS version 2.1 • May be implemented in CASMO5 May 24-25, 2011 International Users Group Meeting International Users Group Meeting September 2009
© Copyright 2026 Paperzz