Package jline.solvers

Class SolverOptions.Config

java.lang.Object

jline.solvers.SolverOptions.Config

Enclosing class:

SolverOptions

public static class SolverOptions.Config
extends Object

Advanced configuration options for specialized solver features. These options control solver-specific behavior and may not be applicable to all solver types.

Field Summary

Fields

Modifier and Type	Field	Description
Integer	aghq_nodes	options.config.aghq_nodes: nodes per simplex direction of the adaptive Gauss-Hermite rule.
double	aoi_preemption	AoI preemption/replacement probability override for Age of Information analysis.
String	compress	State space compression method
int	ctmc_tv_ngrid	CTMC time-varying transient: number of points of the uniform time grid over which the non-homogeneous generator is propagated (one matrix exponential per interval, multiplier frozen at the interval midpoint).
String	da	CTMC decomposition/aggregation method for Env solver.
int	da_iter	Number of iterations for iterative decomposition/aggregation methods (kms, takahashi).
int	dae_maxcov	Fluid 'dae' transient: largest covariance dimension integrated ALONGSIDE the mean.
int	dae_maxstate	Fluid 'dae': largest phase-resolved state the SIMULTANEOUS closure solve is attempted on.
Double	env_alpha	Alpha parameter for Env solver Courtois decomposition (default: 0.015) Controls the coupling threshold for environment state grouping
boolean	eventcache	Enable event caching for SSA
double	fau_delta	CTMC transient by "fau": occupancy below which a state is dropped from the support.
double	fau_epsilon	CTMC transient by "fau": total probability mass the whole horizon may discard.
int	fau_ngrid	CTMC transient by "fau": number of points of the uniform output grid, used when options.timestep is unset.
boolean	fj_warmstart	Resume the fork-join (MMT) fixed point from the iterate retained by the previous runAnalyzer call on the same solver, instead of restarting from GlobalConstants.FineTol.
Boolean	fluid_earlystop	Fluid: stop the ODE window loop once the geometric tail of the iteration says the state is at a fixed point, instead of always running iter_max windows.
String	fork_join	Fork-join handling strategy
int	gmres_restart	Krylov subspace dimension between restarts for the GMRES path in ctmc_solve.
boolean	hide_immediate	Hide immediate transitions from analysis
String	highvar	High-variance class handling strategy
Matrix	init_cov	Fluid 'kp': initial covariance Sigma(0), dim-by-dim in the same layout as kp_init_sol.
Matrix	interlock	Interlock matrix of Franks (1999), Eq.
Matrix	interlock_chain	The same matrix aggregated to the chain basis, set by the AMVA handler for the iteration it is about to run.
boolean	interlocking	Enable interlocking optimization
double[]	kp_init_sol	Fluid 'kp': initial state vector in the KO-PENDER layout -- one offset counter walking the stations in order, an arrival-phase block at each EXT station-class and a service-phase block at every other, with no mass returning to the source.
String	layer_init	SolverLN layer initialization strategy.
String	layering	SolverLN layering strategy.
String	ln_transient	SolverLN transient coupling mode: "coupled" (default) runs the waveform-relaxation coupled layered transient, "decoupled" freezes the inter-layer demands at the converged fixed point.
String	ln_transient_channels	SolverLN coupled transient: which inter-layer coupling channels are injected, "both" (default), "thinkt" (client-delay only) or "callservt" (synchronous-call service only).
int	ln_transient_iter_max	SolverLN coupled transient: maximum number of waveform-relaxation iterations.
double	ln_transient_tol	SolverLN coupled transient: sup-norm tolerance on the queue-length trajectory change between relaxation iterations.
String	map_env	Random-environment fallback for MAP/MMPP/MMAP models on solvers that cannot consume a non-renewal process: "auto" approximates the model through its environment image, "off" rejects it as before.
int	map_env_maxstages	Cap on the number of environment stages, i.e.
String	map_env_method	Environment recombination used by that fallback: "auto", "meanfield" (transient-capable stage solvers only), "dec" (slow-environment limit) or "avg" (fast-environment limit).
Integer	mcmc_batches
Double	mcmc_burnin	Warm-up fraction the MCMC estimator discards before it starts accumulating; null keeps 0.1.
Integer	mdd_maxiter	Maximum sweeps of the MDD level iteration; null keeps the Mdd_mcd default of 500.
Double	mdd_tol	Convergence tolerance of the MDD level iteration (SolverCTMC method 'mdd'); null keeps the Mdd_mcd default of 1e-12.
String	merge	State merging strategy
Matrix[]	moment_cov	Fluid moment closure: per-station coordinate covariance blocks closing the capacity-share RATIO of the PS/FCFS/DPS/GPS branches, which is a separate closure from the min().
int	moment_maxstate	Fluid moment closure: largest phase-resolved state the covariance (Lyapunov) equation is attempted on.
double[]	moment_sigma2	Fluid moment closure: per-station population variance closing the E[min(X,c)] term of the closing drift.
String	multiserver	Multi-server scheduling strategy
List<FluidRateMultiplier.NhppEntry>	nhpp_sched	Fluid closing ODE: non-homogeneous Poisson (NHPP) source intensities to track over the transient, one entry per (station, class).
String	nonmkv	Method for non-Markovian distribution conversion.
int	nonmkvorder	Order (number of phases) for non-Markovian distribution approximation.
String	np_priority	Non-preemptive priority handling
int	num_cdf_pts	Number of points for CDF (Cumulative Distribution Function) computation.
int	orbit_maxlevel	Fixed orbit truncation level for the matrix-analytic retrial solver.
double	orbit_tailtol	Relative orbit-truncation error target of the adaptive refinement in the matrix-analytic retrial solver.
String	passage_method	SolverCTMC.getCdfFirstPassT: "expm" (default) or "lt", selecting the dense matrix-exponential route or the Laplace-transform inversion of ctmc_passage_time.
String	phfit	Family used when a concrete distribution is replaced by a Markovian surrogate: "cme" fits a concentrated matrix exponential plus an exponential tail, "ph" keeps the Erlang/Bernstein phase-type.
boolean	preserveDet	Whether to preserve deterministic distributions during non-Markovian approximation (true keeps Det, false approximates with PH).
List<Double>	pstar	P-norm smoothing parameters for fluid solvers
List<FluidRateMultiplier.RateEntry>	rate_sched	Fluid closing ODE: explicit per-(station,class) rate trajectories.
Matrix	rate_traj_mmat	Fluid closing ODE: event rate multiplier matrix of the caller-supplied trajectory, of size numEvents x numel(rate_traj_tgrid).
double[]	rate_traj_tgrid	Fluid closing ODE: time grid of the caller-supplied event rate multiplier trajectory.
String	relax	Under-relaxation mode for SolverLN convergence improvement.
double	relax_factor	Relaxation factor (omega) when relaxation is enabled.
int	relax_history	Error history window size for adaptive relaxation mode.
double	relax_min	Minimum relaxation factor for adaptive mode.
boolean	remote	Enable remote execution via REST API (LQNS-specific).
String	remote_url	URL of lqns-rest server for remote execution (LQNS-specific).
Boolean	rqna_alpha	RQNA alpha correction terms toggle (eq.
Boolean	rqna_beta	RQNA beta correction terms toggle (eqs.
Boolean	rqna_feedback_elim	RQNA (Robust Queueing Network Analyzer) near-immediate feedback elimination toggle.
Double	rqt_alpha_a	RQT tail coefficient in (1,2] of the external arrival processes.
Double	rqt_alpha_s	RQT tail coefficient in (1,2] of the service processes.
Boolean	rqt_exact	RQT toggle replacing the closed-form bound of Theorem 3 by the exact worst case over the uncertainty sets.
String	rqt_regime	RQT (Robust Queueing Theory) service adaptation regime of Table 1: "independent" (default, service distribution unknown), "normal" or "pareto".
Object	runLengthPlan	`options.config.runLengthPlan`: ask a simulation solver how long its run SHOULD have been for a target relative precision.
Double	sjn_lattice_max	Lattice size above which the shortest-job-next dispatch prefers the Schweitzer fixed point over the population recursion, under method 'default'.
Double	sjn_lfactor	Grid extent at a shortest-job-next station, in units of the largest mean service time.
Integer	sjn_ns	Number of grid subdivisions of the job size axis at a shortest-job-next station, even.
Double	sjn_umax	Utilization cap at a shortest-job-next station, strictly below one.
Double	slotlength	Slot length in model time units, used when slotted is true; null keeps unit slots.
boolean	slotted	Run the solver on a discrete time scale (slot lattice) rather than on the continuous time axis.
int	space_max	Maximum state space size
String	state_space_gen	State space generation strategy
String	stochiter	Stochastic iteration mode for SolverLN when one or more layer solvers return noisy estimates (simulation or Monte Carlo).
double	stochiter_a0	Robbins-Monro initial step a0 applied after burn-in.
double	stochiter_alpha	Robbins-Monro step decay exponent alpha in (0.5,1].
int	stochiter_burnin	Picard burn-in iterations before the Robbins-Monro step decay starts.
int	stochiter_conseq	Consecutive sub-tolerance iterations of the averaged-iterate drift required to stop.
String	symbolic	Computer algebra backend for symbolic analysis: "auto" searches for a line-sage-rest service and starts one from a local image if needed, "none" disables it, and any other value is either a service URL or a Docker image name.
int	symbolic_timeout	Per-request timeout of the symbolic backend, in seconds.
List<int[]>	tbi_cells	Trajectory-based iteration (TBI) fluid method: explicit station partition.
int	tbi_cellsize	Trajectory-based iteration (TBI) fluid method: target number of stations per cell used by the greedy agglomerative partition when tbi_cells is not provided.
int	tbi_iter_max	Trajectory-based iteration (TBI) fluid method: maximum number of waveform relaxation sweeps per time segment.
double	tbi_tol	Trajectory-based iteration (TBI) fluid method: absolute sup-norm tolerance on the state trajectory used to stop the Jacobi waveform relaxation sweeps within a time segment.
String	transient_method	CTMC transient method: "ode" (default) integrates the forward equation, "fau" marches fast adaptive uniformization (Ctmc_fau) over the output grid.
String	variates	Variance reduction technique for LDES simulation.
Double	warmup	Warmup length as an ABSOLUTE number of samples to discard; null leaves the transient filter on warmupfrac.
Double	warmupfrac	SSA warmup discard fraction.

Constructor Summary

Constructors

Constructor	Description
Config()

Method Summary

Modifier and Type	Method	Description
boolean	containsKey(String key)	Check if a configuration parameter exists
Object	get(String key)	Retrieve a configuration parameter
void	put(String key, Object value)	Store a configuration parameter

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Details

slotted

public boolean slotted

Run the solver on a discrete time scale (slot lattice) rather than on the continuous time axis. SolverLDES simulates the lattice directly; SolverNC routes to the discrete-time product-form analyzer (Solver_nc_dt_analyzer) and refuses a model outside it, rather than falling back to a continuous-time approximation. Default: false.

slotlength

public Double slotlength

Slot length in model time units, used when slotted is true; null keeps unit slots. Metrics are computed per slot and rescaled by this factor, so the caller reads back the units it built the model in.

highvar

public String highvar

High-variance class handling strategy

passage_method

public String passage_method

SolverCTMC.getCdfFirstPassT: "expm" (default) or "lt", selecting the dense matrix-exponential route or the Laplace-transform inversion of ctmc_passage_time. Null means the default.

fluid_earlystop

public Boolean fluid_earlystop

Fluid: stop the ODE window loop once the geometric tail of the iteration says the state is at a fixed point, instead of always running iter_max windows. Null means the default, which is on.

warmupfrac

public Double warmupfrac

SSA warmup discard fraction. When set and positive the serial SSA engine discards this fraction of the collected trajectory from the mean estimates (mirroring MATLAB solver_ssa.m) and the batch-means confidence intervals use it as their transient discard; null or 0 disables the mean discard and the CI falls back to its legacy 10%.

warmup

public Double warmup

Warmup length as an ABSOLUTE number of samples to discard; null leaves the transient filter on warmupfrac.

The simulators only take a FRACTION of the run, so a count is meaningful only against options.samples and is converted to warmupfrac = warmup / samples when the run is configured, not when it is parsed: the two may be given in either order. A count at or beyond the sample budget would discard the whole run and is refused rather than clamped.

mdd_tol

public Double mdd_tol

Convergence tolerance of the MDD level iteration (SolverCTMC method 'mdd'); null keeps the Mdd_mcd default of 1e-12.

Kept separate from iter_tol on purpose: that is the solver-level fixed-point tolerance, sized for AMVA outer loops, while the level iteration is an inner numerical solve whose result is checked against the population invariant at 1e-6. Feeding iter_tol here stops the iteration short of the fixed point and trips that guard.

mdd_maxiter

public Integer mdd_maxiter

Maximum sweeps of the MDD level iteration; null keeps the Mdd_mcd default of 500.

aghq_nodes

public Integer aghq_nodes

options.config.aghq_nodes: nodes per simplex direction of the adaptive Gauss-Hermite rule. q = 1 reproduces pfqn_le; the rule costs q^(M-1) evaluations, so the default stays small.

mcmc_batches

public Integer mcmc_batches

mcmc_burnin

public Double mcmc_burnin

Warm-up fraction the MCMC estimator discards before it starts accumulating; null keeps 0.1. The paper discards none and ignores the initialization bias.

multiserver

public String multiserver

Multi-server scheduling strategy

np_priority

public String np_priority

Non-preemptive priority handling

pstar

public List<Double> pstar

P-norm smoothing parameters for fluid solvers

rqna_feedback_elim

public Boolean rqna_feedback_elim

RQNA (Robust Queueing Network Analyzer) near-immediate feedback elimination toggle. Null (default) applies elimination (Whitt-You Algorithm 2 / Section 4.2); set FALSE for plain RQNA (Algorithm 1).

rqna_alpha

public Boolean rqna_alpha

RQNA alpha correction terms toggle (eq. 34). Null (default) = enabled.

rqna_beta

public Boolean rqna_beta

RQNA beta correction terms toggle (eqs. 38-39). Null (default) = enabled.

rqt_regime

public String rqt_regime

RQT (Robust Queueing Theory) service adaptation regime of Table 1: "independent" (default, service distribution unknown), "normal" or "pareto".

rqt_exact

public Boolean rqt_exact

RQT toggle replacing the closed-form bound of Theorem 3 by the exact worst case over the uncertainty sets. Null (default) = closed form.

rqt_alpha_a

public Double rqt_alpha_a

RQT tail coefficient in (1,2] of the external arrival processes. Null (default) = 2, the finite-variance regime.

rqt_alpha_s

public Double rqt_alpha_s

RQT tail coefficient in (1,2] of the service processes. Null (default) = 2, the finite-variance regime.

variates

public String variates

Variance reduction technique for LDES simulation.

Available options:

• "none": No variance reduction (standard simulation)
• "antithetic": Antithetic variates using synchronized 1-U method. Generates paired samples with negative correlation by using antithetic variates to reduce variance through negative correlation.
• "control": Control variates using mean-based correction. Applies post-hoc corrections based on deviation of sampled means from known theoretical means (E[arrival]=1/λ, E[service]=1/μ).
• "both": Combined antithetic and control variates

Default: "none"

env_alpha

public Double env_alpha

Alpha parameter for Env solver Courtois decomposition (default: 0.015) Controls the coupling threshold for environment state grouping

fork_join

public String fork_join

Fork-join handling strategy

sjn_lattice_max

public Double sjn_lattice_max

Lattice size above which the shortest-job-next dispatch prefers the Schweitzer fixed point over the population recursion, under method 'default'. Null keeps the built-in threshold. Mirrors MATLAB/Python options.config.sjn_lattice_max.

sjn_ns

public Integer sjn_ns

Number of grid subdivisions of the job size axis at a shortest-job-next station, even. Null keeps the built-in value. Mirrors MATLAB/Python options.config.sjn_ns.

sjn_lfactor

public Double sjn_lfactor

Grid extent at a shortest-job-next station, in units of the largest mean service time. Null keeps the built-in value. Mirrors MATLAB/Python options.config.sjn_lfactor.

sjn_umax

public Double sjn_umax

Utilization cap at a shortest-job-next station, strictly below one. Null keeps the built-in value. Mirrors MATLAB/Python options.config.sjn_umax.

fj_warmstart

public boolean fj_warmstart

Resume the fork-join (MMT) fixed point from the iterate retained by the previous runAnalyzer call on the same solver, instead of restarting from GlobalConstants.FineTol. Only has an effect under an outer iteration such as SolverLN, which re-solves each layer once per outer iteration.

map_env

public String map_env

Random-environment fallback for MAP/MMPP/MMAP models on solvers that cannot consume a non-renewal process: "auto" approximates the model through its environment image, "off" rejects it as before.

map_env_method

public String map_env_method

Environment recombination used by that fallback: "auto", "meanfield" (transient-capable stage solvers only), "dec" (slow-environment limit) or "avg" (fast-environment limit).

map_env_maxstages

public int map_env_maxstages

Cap on the number of environment stages, i.e. the product of the phase orders.

merge

public String merge

State merging strategy

compress

public String compress

State space compression method

space_max

public int space_max

Maximum state space size

interlocking

public boolean interlocking

Enable interlocking optimization

interlock

public Matrix interlock

Interlock matrix of Franks (1999), Eq. (4.7), CLASS-indexed: interlock(r,s) is the share of the class-s queue that a class-r arrival must not see, because that work was itself caused by the class-r request. Null for every model but the layers of SolverLN.

interlock_chain

public Matrix interlock_chain

The same matrix aggregated to the chain basis, set by the AMVA handler for the iteration it is about to run. Never set from outside the MVA solvers.

eventcache

public boolean eventcache

Enable event caching for SSA

hide_immediate

public boolean hide_immediate

Hide immediate transitions from analysis

state_space_gen

public String state_space_gen

State space generation strategy

nonmkv

public String nonmkv

Method for non-Markovian distribution conversion.

Available options:

• "none": No conversion, keep distributions as-is
• "bernstein": Convert using Bernstein polynomial approximation to phase-type

Default: "bernstein"

nonmkvorder

public int nonmkvorder

Order (number of phases) for non-Markovian distribution approximation. Higher values provide more accurate approximations but increase computational cost.

Default: 20

phfit

public String phfit

Family used when a concrete distribution is replaced by a Markovian surrogate: "cme" fits a concentrated matrix exponential plus an exponential tail, "ph" keeps the Erlang/Bernstein phase-type. At a budget of nonmkvorder phases the ME reaches an SCV of O(1/n^2) where the Erlang stops at 1/n, and it matches the first two moments exactly. SSA, Fluid and JMT must use "ph".

runLengthPlan

public Object runLengthPlan

`options.config.runLengthPlan`: ask a simulation solver how long its run SHOULD have been for a target relative precision. Null leaves the plan uncomputed. A Double is the target relative precision at the run's own confidence level; a Map<String,Double> with keys relprecision and confidence sets both. The plan lands in result.runLengthPlan and is computed by SimRunlength.sim_runlength_plan(jline.util.matrix.Matrix, jline.util.matrix.Matrix, double, double, double).

preserveDet

public boolean preserveDet

Whether to preserve deterministic distributions during non-Markovian approximation (true keeps Det, false approximates with PH).

da

public String da

CTMC decomposition/aggregation method for Env solver.

Available options:

• "courtois": Courtois decomposition (default)
• "kms": Koury-McAllister-Stewart method
• "takahashi": Takahashi's method
• "multi": Multigrid method

Default: "courtois"

da_iter

public int da_iter

Number of iterations for iterative decomposition/aggregation methods (kms, takahashi).

Default: 10

relax

public String relax

Under-relaxation mode for SolverLN convergence improvement.

Available options:

• "auto": Start without relaxation, enable when oscillation detected (default)
• "fixed": Always use relax_factor
• "adaptive": Adjust omega based on error trajectory
• "none": Disable relaxation

Default: "auto"

layer_init

public String layer_init

SolverLN layer initialization strategy. "bound" initializes the layer throughputs from the Majumdar-Woodside robust box bounds before the first solve; null/"none" uses the default (zero) initialization.

layering

public String layering

SolverLN layering strategy. "srvn" (default) builds one submodel per server; "flat"/"squashed" builds a single submodel holding every processor and task. See _kb/06-solver-catalog.md (LN section).

relax_factor

public double relax_factor

Relaxation factor (omega) when relaxation is enabled. Value should be between 0 and 1, where lower values provide more damping.

Default: 0.1

relax_min

public double relax_min

Minimum relaxation factor for adaptive mode.

Default: 0.1

relax_history

public int relax_history

Error history window size for adaptive relaxation mode.

Default: 5

stochiter

public String stochiter

Stochastic iteration mode for SolverLN when one or more layer solvers return noisy estimates (simulation or Monte Carlo).

Available options:

• "auto": enable Robbins-Monro mode when a stochastic layer solver is detected (default)
• "rm": Robbins-Monro step decay with Polyak-Ruppert averaging
• "crn": common random numbers (pinned per-layer seeds), standard convergence test
• "off": deterministic Picard iteration

stochiter_alpha

public double stochiter_alpha

Robbins-Monro step decay exponent alpha in (0.5,1].

Default: 0.6

stochiter_a0

public double stochiter_a0

Robbins-Monro initial step a0 applied after burn-in.

Default: 1.0

stochiter_burnin

public int stochiter_burnin

Picard burn-in iterations before the Robbins-Monro step decay starts.

Default: 5

stochiter_conseq

public int stochiter_conseq

Consecutive sub-tolerance iterations of the averaged-iterate drift required to stop.

Default: 3

num_cdf_pts

public int num_cdf_pts

Number of points for CDF (Cumulative Distribution Function) computation. Used by SolverFluid and SolverMAM for response time distribution analysis.

Default: 200 (100 for MAM solver)

aoi_preemption

public double aoi_preemption

AoI preemption/replacement probability override for Age of Information analysis. Set to a value between 0 and 1 to override automatic detection. NaN means auto-detect from scheduling strategy.

Default: NaN (auto-detect)

remote

public boolean remote

Enable remote execution via REST API (LQNS-specific). When true, solver uses HTTP to communicate with lqns-rest server.

Default: false

remote_url

public String remote_url

URL of lqns-rest server for remote execution (LQNS-specific).

Default: "http://localhost:8080"

symbolic

public String symbolic

Computer algebra backend for symbolic analysis: "auto" searches for a line-sage-rest service and starts one from a local image if needed, "none" disables it, and any other value is either a service URL or a Docker image name. See SymEngines.

Default: "auto"

symbolic_timeout

public int symbolic_timeout

Per-request timeout of the symbolic backend, in seconds. Symbolic solves grow superpolynomially in the number of states, so this bounds both the client wait and the server side computation.

Default: 300

tbi_tol

public double tbi_tol

Trajectory-based iteration (TBI) fluid method: absolute sup-norm tolerance on the state trajectory used to stop the Jacobi waveform relaxation sweeps within a time segment.

Default: 1e-3

tbi_iter_max

public int tbi_iter_max

Trajectory-based iteration (TBI) fluid method: maximum number of waveform relaxation sweeps per time segment.

Default: 50

gmres_restart

public int gmres_restart

Krylov subspace dimension between restarts for the GMRES path in ctmc_solve. Nonpositive leaves the kernel default of min(n, 50).

Default: 0

tbi_cellsize

public int tbi_cellsize

Trajectory-based iteration (TBI) fluid method: target number of stations per cell used by the greedy agglomerative partition when tbi_cells is not provided.

Default: 5

tbi_cells

public List<int[]> tbi_cells

Trajectory-based iteration (TBI) fluid method: explicit station partition. Each int[] is a cell of zero-based station indices; the cells must be a disjoint cover of 0..nstations-1. Null (default) triggers the greedy agglomerative partition.

rate_traj_tgrid

public double[] rate_traj_tgrid

Fluid closing ODE: time grid of the caller-supplied event rate multiplier trajectory. Paired with rate_traj_mmat; both must be set for the channel to be active. Mirrors the MATLAB options.config.rate_traj cell {tgrid, Mmat}.

Default: null (no time-varying multiplier)

rate_traj_mmat

public Matrix rate_traj_mmat

Fluid closing ODE: event rate multiplier matrix of the caller-supplied trajectory, of size numEvents x numel(rate_traj_tgrid). Used by the coupled LN layer transient to inject time-varying inter-layer demand.

Default: null

nhpp_sched

public List<FluidRateMultiplier.NhppEntry> nhpp_sched

Fluid closing ODE: non-homogeneous Poisson (NHPP) source intensities to track over the transient, one entry per (station, class). Injected by SolverFluid.getTranAvg; steady-state analysis leaves this null, the NHPP steady state being its time average. Mirrors the MATLAB options.config.nhpp_sched.

Default: null

rate_sched

public List<FluidRateMultiplier.RateEntry> rate_sched

Fluid closing ODE: explicit per-(station,class) rate trajectories. Mirrors the MATLAB options.config.rate_sched.

Default: null

kp_init_sol

public double[] kp_init_sol

Fluid 'kp': initial state vector in the KO-PENDER layout -- one offset counter walking the stations in order, an arrival-phase block at each EXT station-class and a service-phase block at every other, with no mass returning to the source. Mirrors the MATLAB options.config.kp_init_sol.

NOT SolverOptions.init_sol: that one is laid out for the CLOSING state vector, and consuming it here silently zeroes the source phase mass and with it the whole network. A wrong-sized seed is refused rather than ignored.

Default: null

init_cov

public Matrix init_cov

Fluid 'kp': initial covariance Sigma(0), dim-by-dim in the same layout as kp_init_sol. A caller that carries a DISTRIBUTION across a handoff supplies the second moment beside the mean, so the next stage does not restart from a point mass it never had. Null keeps the default diag(theta) - theta theta' of the initial arrival phase.

Default: null

moment_sigma2

public double[] moment_sigma2

Fluid moment closure: per-station population variance closing the E[min(X,c)] term of the closing drift. Null or all-zero selects the first-order closure, which is what every method other than minnormal uses, so the legacy code path stays bit-identical. Set by the outer fixed point of the moment-closure driver.

Default: null

moment_cov

public Matrix[] moment_cov

Fluid moment closure: per-station coordinate covariance blocks closing the capacity-share RATIO of the PS/FCFS/DPS/GPS branches, which is a separate closure from the min(). One entry per station, null where the station needs none. Paired with moment_sigma2.

Default: null

moment_maxstate

public int moment_maxstate

Fluid moment closure: largest phase-resolved state the covariance (Lyapunov) equation is attempted on. The solve is cubic in this dimension and the covariance is dense, so the driver refuses rather than silently crawling above it.

Default: 200

dae_maxstate

public int dae_maxstate

Fluid 'dae': largest phase-resolved state the SIMULTANEOUS closure solve is attempted on. Lower than moment_maxstate because this route takes a finite-difference Jacobian over the unknowns rather than solving one Lyapunov equation, so its cost is quartic and not cubic. Above it the driver refuses and names options.method='minnormal', which computes the same closure by successive substitution.

Default: 100

dae_maxcov

public int dae_maxcov

Fluid 'dae' transient: largest covariance dimension integrated ALONGSIDE the mean. The covariance adds nc^2 differential states and the Jacobian is formed by finite differences over all of them, so the cost grows as nc^4. Above this the mean is still integrated as a DAE -- population conservation stays an algebraic equation -- but the variance is held at its stationary value, which is what 'minnormal' does for the whole of its transient anyway.

Default: 25

ctmc_tv_ngrid

public int ctmc_tv_ngrid

CTMC time-varying transient: number of points of the uniform time grid over which the non-homogeneous generator is propagated (one matrix exponential per interval, multiplier frozen at the interval midpoint). Only read when rate_sched is set. Mirrors the MATLAB options.config.ctmc_tv_ngrid.

Default: 100

transient_method

public String transient_method

CTMC transient method: "ode" (default) integrates the forward equation, "fau" marches fast adaptive uniformization (Ctmc_fau) over the output grid. It is a config key rather than a method name because it changes no stationary answer, and because a new entry in the solver's valid-method list is enumerated by the sanity harness, which then wants a baseline per method. Mirrors the MATLAB options.config.transient_method.

Default: "ode"

fau_epsilon

public double fau_epsilon

CTMC transient by "fau": total probability mass the whole horizon may discard. It is divided by the number of grid steps, each step removing mass and none putting any back, so the accumulated defect stays below this. Mirrors options.config.fau_epsilon.

Default: 1e-6

fau_delta

public double fau_delta

CTMC transient by "fau": occupancy below which a state is dropped from the support. Mirrors options.config.fau_delta.

Default: 1e-12

fau_ngrid

public int fau_ngrid

CTMC transient by "fau": number of points of the uniform output grid, used when options.timestep is unset. Mirrors options.config.fau_ngrid.

Default: 100

ln_transient

public String ln_transient

SolverLN transient coupling mode: "coupled" (default) runs the waveform-relaxation coupled layered transient, "decoupled" freezes the inter-layer demands at the converged fixed point. Mirrors the MATLAB options.config.ln_transient.

ln_transient_channels

public String ln_transient_channels

SolverLN coupled transient: which inter-layer coupling channels are injected, "both" (default), "thinkt" (client-delay only) or "callservt" (synchronous-call service only). Mirrors the MATLAB options.config.ln_transient_channels.

ln_transient_iter_max

public int ln_transient_iter_max

SolverLN coupled transient: maximum number of waveform-relaxation iterations. Mirrors options.config.ln_transient_iter_max.

Default: 20

ln_transient_tol

public double ln_transient_tol

SolverLN coupled transient: sup-norm tolerance on the queue-length trajectory change between relaxation iterations. Mirrors options.config.ln_transient_tol.

Default: 1e-2

orbit_maxlevel

public int orbit_maxlevel

Fixed orbit truncation level for the matrix-analytic retrial solver. Non-positive (default) selects the adaptive, residual-driven level.

Default: 0

orbit_tailtol

public double orbit_tailtol

Relative orbit-truncation error target of the adaptive refinement in the matrix-analytic retrial solver.

Default: 1e-6

Constructor Details

Config

public Config()

Method Details

put

public void put(String key,
 Object value)

Store a configuration parameter

get

public Object get(String key)

Retrieve a configuration parameter

containsKey

public boolean containsKey(String key)

Check if a configuration parameter exists