odeMethodToInt: Conversion between character and integer ODE integration...
In rxode2: Facilities for Simulating from ODE-Based Models

odeMethodToInt

R Documentation

Conversion between character and integer ODE integration methods for rxode2

Description

If NULL is given as the method, all choices are returned as a named vector.

Usage

odeMethodToInt(
  method = c("liblsoda", "lsoda", "dop853", "indLin", "f78", "rk4", "ck54", "ab", "abm",
    "dop5", "bs", "ros4", "iem", "sem", "sb3a", "sb3am4", "vv", "mm", "em", "cvode",
    "trapz", "ssp3", "f32", "rk43", "dop54", "vern65", "vern76", "dop87", "vern98",
    "ros43", "ros6", "backwardEuler", "gauss6", "iiic6", "radauiia5", "geng5", "sdirk43",
    "euler", "midpoint", "heun", "ssp22", "rk3", "ssp53", "s4", "r4", "ls44", "ls54",
    "ssp54", "s5", "rk5", "c5", "l5", "lk5a", "lk5b", "b6", "s7", "s8_10", "cv8",
    "s8_12", "s10", 
     "z10", "o10", "h10", "dp54", "v65e", "v76e", "dp87", "v98e",
    "ssp33", "bs32", "ssp43", "f45", "t54", "s54", "pp54", "pp54b", "bs54", "ss54",
    "dp65", "c65", "tp64", "v65r", "v65", "dverk65", "tf65", "tp75", "tmy7", "tmy7s",
    "v76r", "ss76", "v78", "dverk78", "dp85", "tp86", "v87e", "v87r", "ev87", "k87",
    "f89", "v89", "t98a", "v98r", "s98", "f108", "c108", "b109", "s1110a", "f1210",
    "o129", "f1412", "lsode", "bdf", "rk4s", "eulers", "midpoints", "heuns", "dop5s",
    "dop853s", "ck54s", "bs32s", "vern65s", 
     "vern76s", "dop87s", "f78s", "ros4s",
    "radauiia5s", "backwardEulers", "gauss6s", "sdirk43s", "iiic6s", "ros43s", "ros6s",
    "geng5s", "rk3s", "rk43s", "cvodesadj", "liblsodaadj", "abs", "dop54s", "dp54s",
    "vern98s", "f45s", "t54s", "pp54s", "pp54bs", "bs54s", "ss54s", "dp65s", "c65s",
    "tp64s", "v65rs", "dverk65s", "tf65s", "tp75s", "tmy7sadj", "tmy7adj", "v76rs",
    "ss76s", "v78s", "dverk78s", "dp85s", "tp86s", "v87es", "v87rs", "ev87s", "k87s",
    "v89s", "t98as", "v98rs", "s98s", "c108s", "b109s", 
     "s1110as", "o129s")
)

Arguments

method

The method for solving ODEs. Currently this supports:

"liblsoda" thread safe lsoda. This supports parallel thread-based solving, and ignores user Jacobian specification.
"lsoda" – LSODA solver. Does not support parallel thread-based solving, but allows user Jacobian specification.
"dop853" – DOP853 solver. Does not support parallel thread-based solving nor user Jacobian specification
"indLin" – Solving through inductive linearization. The rxode2 dll must be setup specially to use this solving routine.
"f78" – Runge-Kutta Fehlberg 78 solver using Boost's odeint library.
"rk4" – Runge-Kutta 4 solver using Boost's odeint library.
"ck54" – Cash-Karp 5(4) solver using Boost's odeint library.
"ab" – Adams-Bashforth multi-step solver with a direct implementation (order 1-8, default 5). Uses RK4 to initialize the derivative history and then applies the Adams-Bashforth extrapolation formula. Is a fixed-step method (step size controlled by hmin).
"abm" – Adams-Bashforth-Moulton solver using Boost's odeint library.
"dop5" – DOPRI5 solver using Boost's odeint library (supports dense output).
"bs" – Bulirsch-Stoer solver using Boost's odeint library (supports dense output).
"ros4" (implicit) – Rosenbrock 4 solver using Boost's odeint library (supports dense output). Requires an analytical Jacobian (auto-computed from the model when not provided); falls back to liblsoda if Jacobian generation fails.
"ros43" (implicit) – Kaps-Rentrop 4th-order A-stable Rosenbrock method (GRK4A) from libode. Adaptive step size via embedded 3rd-order error estimate. Requires an analytical Jacobian (auto-computed from the model when not provided); falls back to liblsoda if Jacobian generation fails.
"ros6" (implicit) – Kaps-Wanner 6th-order A-stable Rosenbrock method (ROW6A) from libode. Fixed step size (set with hmin). Requires Jacobian; falls back to liblsoda if unavailable.
"backwardEuler" (implicit) – Backward Euler (BDF-1), 1st-order L-stable fully implicit method from libode. Fixed step size (set with hmin). Requires Jacobian; falls back to liblsoda if unavailable.
"gauss6" (implicit) – Gauss-Legendre 6th-order A-stable fully-implicit method (3 stages) from libode. Fixed step size. Requires Jacobian; falls back to liblsoda if unavailable.
"iiic6" (implicit) – Lobatto IIIC 6th-order L-stable fully-implicit method (4 stages) from libode. Fixed step size. Requires Jacobian; falls back to liblsoda if unavailable.
"radauiia5" (implicit) – Radau IIA 5th-order L-stable fully-implicit method (3 stages) from libode. Fixed step size (set with hmin). Requires Jacobian; falls back to liblsoda if unavailable.
"geng5" (implicit) – Geng 5th-order symplectic fully-implicit method (3 stages) from libode. Fixed step size. Requires Jacobian; falls back to liblsoda if unavailable.
"sdirk43" (implicit) – L-stable SDIRK 4(3) pair from libode. Adaptive step size via embedded 3rd-order error estimate. Requires Jacobian; falls back to liblsoda if unavailable.
"iem" (implicit) – Implicit Euler solver using Boost's odeint library. Requires an explicit Jacobian (auto-generated via calcJac=TRUE when not provided). Uses Boost.uBLAS vectors and is a fixed-step method (step size controlled by hmin).
"sem" – Symplectic Euler solver using Boost's odeint library. Requires splitting the state into coordinate-momentum pairs (and automatically pads odd-dimensional systems). Is a fixed-step method (step size controlled by hmin).
"sb3a" – Symplectic Runge-Kutta-Nystrom SB3A McLachlan stepper using Boost's odeint library. Requires splitting the state into coordinate-momentum pairs (and automatically pads odd-dimensional systems). Is a fixed-step method (step size controlled by hmin).
"sb3am4" – Symplectic Runge-Kutta-Nystrom SB3A M4 McLachlan stepper using Boost's odeint library. Requires splitting the state into coordinate-momentum pairs (and automatically pads odd-dimensional systems). Is a fixed-step method (step size controlled by hmin).
"vv" – Velocity Verlet stepper using Boost's odeint library. Requires splitting the state into coordinate-momentum pairs (and automatically pads odd-dimensional systems). Is a fixed-step method (step size controlled by hmin).
"mm" – Modified Midpoint stepper using Boost's odeint library. Is a fixed-step method (step size controlled by hmin) and uses the order parameter to configure the number of intermediate steps.
"em" – Explicit Euler stepper using Boost's odeint library. Is a fixed-step method (step size controlled by hmin).
"cvode" – CVODE BDF stiff solver from the SUNDIALS library (sources and headers vendored into rxode2). Supports thread-parallel solving and per-compartment absolute tolerances.
"trapz" – Explicit trapezoidal rule (Heun's method), a 2nd-order fixed-step Runge-Kutta method implemented via the libode library. Each inter-event interval is integrated with fixed steps of size hmin (default 0.01 when hmin=0). If hmin would require more than maxsteps steps to span the interval, the step size is automatically enlarged to stay within that limit. When an inter-event interval is shorter than the nominal step size (e.g., two doses very close together), the step is silently clamped to the interval length so that short intervals are always handled correctly. Supports parallel thread-based solving and steady-state (ss=1) dosing. The method does not use atol or rtol (fixed-step; no error control). Steady-state convergence is governed by ssAtol, ssRtol, minSS, maxSS, and strictSS.
"ssp3" – Strong Stability-Preserving Runge-Kutta of order 3 (SSP-RK3, Shu-Osher method), implemented via the libode library. A 3-stage, 3rd-order explicit fixed-step method with superior non-oscillatory properties for problems with discontinuities or sharp fronts. Uses the Butcher tableau c2=1, a21=1; c3=1/2, a31=1/4, a32=1/4; b1=1/6, b2=1/6, b3=2/3. The step-size behavior, clamping, and steady-state options (hmin, maxsteps, ssAtol, ssRtol, minSS, maxSS, strictSS) are identical to "trapz". Does not use atol or rtol (fixed-step; no error control). Supports parallel thread-based solving and steady-state (ss=1) dosing.
"f32" – Fehlberg 3(2) embedded pair from libode (class OdeRKF32). A 3-stage, 3rd-order adaptive method with a built-in 2nd-order error estimate for automatic step-size control. Tableau: c2=1, a21=1; c3=1/2, a31=1/4, a32=1/4; primary (3rd-order) weights d1=1/6, d2=1/6, d3=4/6; embedded (2nd-order) weights b1=1/2, b2=1/2. Uses atol and rtol for error control. The hmin parameter sets the initial step size (default 0.01); subsequent steps are chosen adaptively. The total number of accepted and rejected steps is bounded by maxsteps. Supports parallel thread-based solving and steady-state (ss=1) dosing with convergence governed by ssAtol, ssRtol, minSS, maxSS, and strictSS.
"rk43" – Runge-Kutta 4(3) pair from libode. A 5-stage, 4th-order adaptive method with a built-in 3rd-order embedded error estimate for automatic step-size control. Tableau: c2=1/3, a21=1/3; c3=2/3, a31=-1/3, a32=1; c4=1, a41=1, a42=-1, a43=1; a5j=bj (FSAL). Primary (4th-order) weights b1=b4=1/8, b2=b3=3/8; embedded (3rd-order) weights d1=1/12, d2=1/2, d3=1/4, d5=1/6. Since a5j=bj the 5th stage evaluation is reused as the 1st stage of the next step (FSAL). Uses atol and rtol for error control. The hmin parameter sets the initial step size (default 0.01); subsequent steps are chosen adaptively. The total number of steps is bounded by maxsteps. Supports parallel thread-based solving and steady-state (ss=1) dosing with convergence governed by ssAtol, ssRtol, minSS, maxSS, and strictSS.
"dop54" – Dormand-Prince 5(4) FSAL method (Dormand & Prince 1980), implemented via the libode library. The same algorithm as MATLAB's ode45. A 7-stage, 5th-order adaptive method with an embedded 4th-order error estimate for automatic step-size control (FSAL: the 7th stage evaluation is reused as the 1st stage of the next step, giving effectively 6 function evaluations per accepted step). Uses atol and rtol for error control. The hmin parameter sets the initial step size (default 0.01); subsequent steps are chosen adaptively. The total number of steps is bounded by maxsteps. Supports parallel thread-based solving and steady-state (ss=1) dosing with convergence governed by ssAtol, ssRtol, minSS, maxSS, and strictSS. NaN/Inf in derivatives (e.g. from NA parameters) is detected immediately and the solve exits with NA output for the affected subject.
"vern65" – Jim Verner's "most efficient" 6(5) FSAL pair (9 stages), implemented via the libode library using coefficients from Verner's own website (RKV65.IIIXb.Efficient). A 6th-order adaptive method with an embedded 5th-order error estimate. The sparse tableau (many zero a-coefficients) reduces function evaluations; the FSAL property means the 9th stage evaluation is reused as the 1st stage of the next step, giving effectively 8 function evaluations per accepted step. Uses atol and rtol for error control. The hmin parameter sets the initial step size (default 0.01); subsequent steps are chosen adaptively. The total number of steps is bounded by maxsteps. Supports parallel thread-based solving and steady-state (ss=1) dosing with convergence governed by ssAtol, ssRtol, minSS, maxSS, and strictSS. NaN/Inf in derivatives is detected immediately and the solve exits with NA output.
"vern76" – Jim Verner's "most efficient" 7(6) pair (10 stages), implemented via the libode library using coefficients from Verner's own website (RKV76.IIa.Efficient). A 7th-order adaptive method with an embedded 6th-order error estimate. The sparse tableau reduces function evaluations per step. Uses atol and rtol for error control. The hmin parameter sets the initial step size (default 0.01); subsequent steps are chosen adaptively. The total number of steps is bounded by maxsteps. Supports parallel thread-based solving and steady-state (ss=1) dosing with convergence governed by ssAtol, ssRtol, minSS, maxSS, and strictSS. NaN/Inf in derivatives is detected immediately and the solve exits with NA output.
"dop87" – Dormand-Prince 8(7) pair (13 stages), implemented via the libode library using coefficients from Hairer, Norsett and Wanner (1993) "Solving ODEs I" (2nd ed.). An 8th-order adaptive method with an embedded 7th-order error estimate. Both solutions are computed from the original state in the final step loop. Uses atol and rtol for error control. The hmin parameter sets the initial step size (default 0.01); subsequent steps are chosen adaptively. The total number of steps is bounded by maxsteps. Supports parallel thread-based solving and steady-state (ss=1) dosing with convergence governed by ssAtol, ssRtol, minSS, maxSS, and strictSS. NaN/Inf in derivatives is detected immediately and the solve exits with NA output.
"vern98" – Jim Verner's "most efficient" 9(8) pair (16 stages), implemented via the libode library using coefficients from Verner's own website (RKV98.IIa.Efficient). A 9th-order adaptive method with an embedded 8th-order error estimate. The highly sparse tableau (stages 8-16 use only k_{0} and k_{5..}) minimizes function evaluations per step. Both solutions are computed from the original state in the final step loop. Uses atol and rtol for error control. The hmin parameter sets the initial step size (default 0.01); subsequent steps are chosen adaptively. The total number of steps is bounded by maxsteps. Supports parallel thread-based solving and steady-state (ss=1) dosing with convergence governed by ssAtol, ssRtol, minSS, maxSS, and strictSS. NaN/Inf in derivatives is detected immediately and the solve exits with NA output.

Aliases for existing methods (no additional C++ code; hmin, atol, rtol, and maxsteps follow the aliased method):

"dp54" – alias for "dop54" (Dormand-Prince 5(4) FSAL, libode).
"v65e" – alias for "vern65" (Verner 6(5) efficient, libode).
"v76e" – alias for "vern76" (Verner 7(6) efficient, libode).
"dp87" – alias for "dop87" (Dormand-Prince 8(7), libode).
"v98e" – alias for "vern98" (Verner 9(8) efficient, libode).
"ssp33" – alias for "ssp3" (Strong Stability-Preserving RK3, libode).

libode fixed-step explicit methods. All use hmin as the fixed step size (default 0.01 when hmin=0); atol and rtol are ignored (no error control); maxsteps bounds the total number of steps; NaN/Inf in derivatives sets ind$err and the subject exits with NA output. All support parallel thread-based solving.

"euler" – Forward (explicit) Euler, 1st-order, 1 stage. Requires a very small step size for accuracy (e.g., hmin=1e-4); intended for pedagogical use or method-comparison baselines.
"midpoint" – Explicit midpoint rule, 2nd-order, 2 stages.
"heun" – Heun's method (explicit trapezoid), 2nd-order, 2 stages. Identical in structure to "trapz" but uses the libode fixed-step driver.
"ssp22" – Strong Stability-Preserving 2-stage 2nd-order method (SSP-RK22), 2 stages. Superior non-oscillatory properties near discontinuities.
"rk3" – Classical 3rd-order Runge-Kutta (Kutta 1901), 3 stages.
"ssp53" – Strong Stability-Preserving 5-stage 3rd-order method (SSP-RK53), 5 stages. High SSP coefficient for hyperbolic PDEs or event-heavy ODE systems.
"s4" – Shanks 4th-order method, 4 stages.
"r4" – Ralston's 4th-order method, 4 stages. Minimizes local truncation error among classical 4-stage 4th-order methods.
"ls44" – Low-storage 4th-order method, 4 stages. Uses a 2-register update scheme that minimizes memory bandwidth at the cost of a less general tableau structure.
"ls54" – Low-storage 4th-order method, 5 stages. Five-stage variant of the 2-register low-storage scheme.
"ssp54" – Strong Stability-Preserving 5-stage 4th-order method (SSP-RK54), 5 stages.
"s5" – Shanks 5th-order method, 5 stages.
"rk5" – Classical 5th-order Runge-Kutta, 6 stages.
"c5" – Cassity 5th-order method, 6 stages.
"l5" – Lawson 5th-order method, 6 stages.
"lk5a" – Luther-Konen 5th-order method, variant A, 6 stages.
"lk5b" – Luther-Konen 5th-order method, variant B, 6 stages.
"b6" – Butcher 6th-order method, 7 stages.
"s7" – Shanks 7th-order method, 9 stages.
"s8_10" – Shanks 8th-order method, 10 stages.
"cv8" – Cooper-Verner 8th-order method, 11 stages.
"s8_12" – Shanks 8th-order method, 12 stages.
"s10" – Stepanov 10th-order method, 15 stages. Requires a moderate step size (e.g., hmin=1.0) because a single step is accurate to very high order.
"z10" – Zhang 10th-order method, 16 stages.
"o10" – Ono 10th-order method, 17 stages.
"h10" – Hairer 10th-order method, 17 stages.

libode adaptive (variable-step) explicit methods. All use atol and rtol for error control; hmin sets the initial step size (default 0.01 when hmin=0); hmax sets the maximum step size; maxsteps bounds total steps; NaN/Inf in derivatives sets ind$err and the subject exits with NA output. All support parallel thread-based solving and steady-state (ss=1) dosing (convergence governed by ssAtol, ssRtol, minSS, maxSS, strictSS).

"bs32" – Bogacki-Shampine 3(2) FSAL pair, 4 stages (Bogacki & Shampine 1989). 3rd-order primary with 2nd-order embedded error estimate. FSAL: the 4th-stage evaluation is reused as the 1st stage of the next step. The same algorithm as Julia BS3() and MATLAB ode23.
"ssp43" – Strong Stability-Preserving 4(3) pair, 4 stages. Adaptive SSP method with a 3rd-order embedded error estimate.
"f45" – Fehlberg 4(5) pair, 6 stages (Fehlberg 1970). 4th-order primary solution with a 5th-order embedded estimate used for error control.
"t54" – Tsitouras 5(4) FSAL pair, 7 stages (Tsitouras 2011). 5th-order primary with 4th-order embedded error estimate. FSAL: the 7th stage is reused as the 1st stage of the next step. The same Butcher tableau as Julia's Tsit5().
"s54" – Stepanov 5(4) FSAL pair, 7 stages. 5th-order primary with 4th-order embedded error estimate.
"pp54" – Papakostas-Papageorgiou 5(4) FSAL pair, 7 stages.
"pp54b" – Papakostas-Papageorgiou 5(4) variant B FSAL pair, 7 stages.
"bs54" – Bogacki-Shampine 5(4) pair, 8 stages. 5th-order primary with 4th-order embedded error estimate (non-FSAL).
"ss54" – Sharp-Smart 5(4) pair, 7 stages.
"dp65" – Dormand-Prince 6(5) pair, 8 stages. 6th-order primary with 5th-order embedded error estimate.
"c65" – Calvo 6(5) pair, 9 stages.
"tp64" – Tsitouras-Papakostas 6(4) pair, 7 stages. 6th-order primary with 4th-order embedded error estimate.
"v65r" – Verner "robust" 6(5) FSAL pair, 9 stages. Robust variant of Verner's 6(5) family with wider stability region.
"v65" – Verner 6(5) pair, 8 stages (non-FSAL).
"dverk65" – Verner DVERK 6(5) pair, 8 stages. Coefficients from the classic DVERK Fortran code distributed by Hull and Enright.
"tf65" – Tsitouras-Famelis 6(5) FSAL pair, 9 stages.
"tp75" – Tsitouras-Papakostas 7(5) pair, 9 stages. 7th-order primary with 5th-order embedded error estimate.
"tmy7" – Tanaka-Muramatsu-Yamashita 7th-order pair, 10 stages. The same family as Julia's TanYam7().
"tmy7s" – Tanaka-Muramatsu-Yamashita 7th-order stable variant, 10 stages. Alternative coefficient set with wider stability region.
"v76r" – Verner "robust" 7(6) pair, 10 stages.
"ss76" – Sharp-Smart 7(6) pair, 11 stages.
"v78" – Verner 7(8) pair, 13 stages. 7th-order primary with 8th-order embedded error estimate. Closest rxode2 analog to Julia Vern8().
"dverk78" – Verner DVERK 7(8) pair, 13 stages. Coefficients from the classic DVERK Fortran code; companion to "dverk65". Also close to Julia Vern8().
"dp85" – Dormand-Prince 8(5) pair, 12 stages. 8th-order primary with 5th-order embedded error estimate.
"tp86" – Tsitouras-Papakostas 8(6) pair, 12 stages. The same family as Julia's TsitPap8().
"v87e" – Verner "efficient" 8(7) pair, 13 stages.
"v87r" – Verner "robust" 8(7) pair, 13 stages.
"ev87" – Enright-Verner 8(7) pair, 13 stages.
"k87" – Kovalnogov-Fedorov-Karpukhina-Simos 8(7) pair, 13 stages.
"f89" – Fehlberg 8(9) pair, 17 stages. 8th-order primary with 9th-order embedded estimate. Same family as MATLAB ode89.
"v89" – Verner 8(9) pair, 16 stages. Alternative to "f89" in the same order bracket. Also close to MATLAB ode89.
"t98a" – Tsitouras 9(8) variant A pair, 16 stages.
"v98r" – Verner "robust" 9(8) pair, 16 stages.
"s98" – Sharp 9(8) pair, 16 stages.
"f108" – Feagin 10(8) pair, 17 stages (Feagin 2007). 10th-order primary with 8th-order embedded error estimate. The same method as Julia's Feagin10(). The large number of stages carries elevated transcription-error risk; verified against Williams' libode canonical order test.
"c108" – Curtis 10(8) pair, 21 stages.
"b109" – Baker 10(9) pair, 21 stages.
"s1110a" – Stone 11(10) variant A pair, 26 stages. 11th-order primary; very few published references.
"f1210" – Feagin 12(10) pair, 25 stages (Feagin 2007). 12th-order primary with 10th-order embedded error estimate. The same method as Julia's Feagin12(). Elevated transcription-error risk due to many stages; verified against Williams' libode canonical order test.
"o129" – Ono 12(9) pair, 29 stages.
"f1412" – Feagin 14(12) pair, 35 stages (Feagin 2007). 14th-order primary with 12th-order embedded error estimate. The same method as Julia's Feagin14(). The highest-order method in rxode2; elevated transcription-error risk due to 35 stages; verified against Williams' libode canonical order test.

deSolve-derived Fortran solvers (from the deSolve R package). Both use non-reentrant Fortran COMMON blocks and therefore run single-threaded only, regardless of the cores setting. They are BDF (Backward Differentiation Formula) stiff solvers and are most useful when you want behavior comparable to deSolve's lsode() or bdf()/vode() functions.

"lsode" – DLSODE (Hindmarsh 1983, ODEPACK) in Adams (non-stiff) mode, MF=10. Variable-order Adams method, order 1-12; no Jacobian evaluation. Corresponds to deSolve's lsode(method="adams"). Adaptive step-size; error controlled by atol and rtol. Not thread-safe – always runs on a single core.
"bdf" – DLSODE (Hindmarsh 1983, ODEPACK) in BDF (stiff) mode, MF=22. Variable-order BDF method, order 1-5; internally generated dense Jacobian. Corresponds to deSolve's bdf() / vode(method="bdf"). Adaptive step-size; error controlled by atol and rtol. Not thread-safe – always runs on a single core.