Transient Analysis Options¶

VAJAX supports VACASK-compatible options for controlling transient analysis behavior. Options can be specified in the netlist and are automatically used as defaults.

Simplest Usage¶

from vajax import CircuitEngine

engine = CircuitEngine('circuit.sim')
engine.parse()
engine.prepare()

# Run with all defaults from netlist - no args needed!
result = engine.run_transient()

The run_transient() method uses: - stop from netlist analysis line as t_stop - step from netlist analysis line as dt - tran_lteratio, nr_convtol, etc. from netlist options

Quick Start¶

Options in netlist are automatically used:

// In your .sim file
options tran_lteratio=1.5 nr_convtol=1.0 reltol=1e-4
analysis top tran step=1p stop=10n

from vajax import CircuitEngine

engine = CircuitEngine('circuit.sim')
engine.parse()

# prepare() automatically uses netlist options
engine.prepare(use_sparse=True)
# engine.adaptive_config.lte_ratio == 1.5  (from tran_lteratio)

Supported Options¶

LTE (Local Truncation Error) Control¶

Netlist Option	AdaptiveConfig	Default	Description
`tran_lteratio`	`lte_ratio`	3.5	LTE tolerance multiplier. Higher values allow larger LTE before reducing timestep.
`tran_redofactor`	`redo_factor`	2.5	Rejection threshold. If `current_dt / new_dt > redo_factor`, the step is rejected and retried.

Tolerances¶

Netlist Option	AdaptiveConfig	Default	Description
`reltol`	`reltol`	1e-3	Relative tolerance for LTE and NR convergence.
`abstol`	`abstol`	1e-12	Absolute tolerance (current) for LTE and NR convergence.
`nr_convtol`	`nr_convtol`	1.0	NR convergence tolerance factor (multiplier on abstol). Values < 1.0 are stricter.

Timestep Control¶

Netlist Option	AdaptiveConfig	Default	Description
`tran_method`	`integration_method`	`trap`	Integration method: `be` (backward Euler), `trap` (trapezoidal), `gear2` (Gear order 2).
`tran_fs`	`tran_fs`	0.25	Initial timestep scale factor. Scales user-specified `step` to start smaller.
`tran_minpts`	`tran_minpts`	50	Minimum output points. Automatically caps `max_dt` to `t_stop / tran_minpts`.
`maxstep`	`max_dt`	inf	Maximum allowed timestep (from tran analysis line).

Convergence Aids¶

Netlist Option	AdaptiveConfig	Default	Description
`tran_gshunt`	`gshunt_init`	0.0	Initial conductance to ground for all nodes. Helps with singular matrices in UIC mode.
-	`gshunt_steps`	5	Steps to ramp gshunt from `gshunt_init` to `gshunt_target`.
-	`gshunt_target`	0.0	Final gshunt value after ramping.

Debug Options¶

AdaptiveConfig	Default	Description
`debug_steps`	False	Print per-step info (time, dt, NR iterations, LTE).
`debug_lte`	False	Print detailed LTE debug info (top contributing nodes).
`progress_interval`	100	Report progress every N steps. Set to 0 to disable.

Internal Parameters¶

AdaptiveConfig	Default	Description
`min_dt`	1e-18	Minimum allowed timestep (seconds).
`max_order`	2	Maximum polynomial order for predictor (0=constant, 1=linear, 2=quadratic).
`grow_factor`	2.0	Maximum factor by which timestep can grow per step.
`warmup_steps`	2	Fixed-dt steps before enabling LTE control (need history for predictor).
`max_consecutive_rejects`	5	Force accept after this many consecutive LTE rejects.

Default Behavior¶

When no explicit AdaptiveConfig is passed to engine.prepare():

Netlist options are parsed during engine.parse() and stored in engine.analysis_params
prepare() builds config from analysis_params automatically
Defaults apply for any options not specified in the netlist

Example with defaults:

# No options in netlist - uses all defaults
engine.prepare()
# lte_ratio=3.5, redo_factor=2.5, reltol=1e-3, etc.

Programmatic Override¶

Explicit AdaptiveConfig completely overrides netlist options:

from vajax.analysis.transient import AdaptiveConfig

# Override specific options
config = AdaptiveConfig(
    lte_ratio=2.0,      # Stricter than default 3.5
    tran_fs=0.1,        # Even smaller initial timesteps
    debug_steps=True,   # Enable per-step debugging
)
engine.prepare(adaptive_config=config)

VACASK Compatibility¶

Supported (Functional)¶

VACASK Option	Status	Notes
`tran_lteratio`	✅ Full	Maps to `lte_ratio`
`tran_redofactor`	✅ Full	Maps to `redo_factor`
`tran_method`	✅ Full	Supports be, trap, gear2
`tran_fs`	✅ Full	Initial timestep scaling
`tran_minpts`	✅ Full	Auto-computes max_dt
`reltol`	✅ Full	Relative tolerance
`abstol`	✅ Full	Absolute tolerance
`nr_convtol`	✅ Full	NR tolerance factor

Not Yet Implemented¶

VACASK Option	Status	Notes
`nr_bypass`	❌	Device bypass for unchanged nodes
`nr_bypasstol`	❌	Bypass tolerance threshold
`nr_contbypass`	❌	Bypass in continuation mode
`tran_predictor`	❌	We always use predictor
`tran_maxord`	Partial	Fixed at order 2
`tran_itl`	❌	Max NR iterations per step
`tran_ft`	❌	Timestep cut factor on NR failure
`tran_xmu`	❌	Trap/Euler mixture parameter

Example: C6288 Benchmark¶

The c6288 multiplier benchmark uses these options:

options nr_convtol=1 tran_lteratio=1.5
analysis top tran step=2p stop=2n icmode="uic"

This sets: - nr_convtol=1.0 - Standard NR tolerance - tran_lteratio=1.5 - Tighter LTE control than default (3.5) - icmode="uic" - Use Initial Conditions (skip DC operating point)

engine = CircuitEngine('c6288.sim')
engine.parse()

# These options are automatically applied
engine.prepare(use_sparse=True)
assert engine.adaptive_config.lte_ratio == 1.5
assert engine.adaptive_config.nr_convtol == 1.0