#818 pwr(V(X,Y),z) out of range error being resolved by adding superfluous computations.

[Josef Salzmann]

ngspice version: 45.2
OS: Linux

An error of pwr(V(X,Y),z) being out of range is somehow resolved by adding superfluous computations to the second argument of pwr.

Example:

.title pwr argument bug
VIN VIN 0 1.0
VCONTROL VCONTROL 0 1.0

X_VRES VIN 0 VCONTROL 0  VC_RES
.SUBCKT VC_RES 1 2 3 4
    R_VAL 1 2 VALUE = { pwr(V(3,4), 0.685841) }
.ENDS

.control
    pre_set strict_errorhandling
    set filetype = ASCII
    save i(vin)

    dc VCONTROL 0.1 1.0 0.1
    write pwr_bug.csv
    exit
.endc
.end

ngspice reports:

Error: 0, -0.314159 out of range for pow

Which is the case when taking the derivative of pwr(V(X,Y),z) for V(X,Y)=0. Since there is a division by 0 when taking 0 to a negative exponent.

However, the simulation finishes without errors when a superfluous computation like "+0" or "*1" is added to the second argument, e.g.:

 R_VAL 1 2 VALUE = { pwr(V(3,4), 0.685841*1) }

How does adding "*1" stop the computation of the derivative?

[Giles Atkinson]

An interesting test case!

The expression parser generates expressions for the differentials of the primary expression with respect to the circuit variables. For pow(a, b), calculating pow(a, b -1) is a shortcut used only when b is a constant. When b is an expression a different formula is used (D(pow(a,b)) = pow(a,b) * (D(b)log(abs(a)) + bD(a)/a)).

So if there is a bug here, it is failing to check the constant value. With that added, no failure. Your test case also shows that constant-valued expressions are not evaluated at parse time, perhaps a useful optimisation.