multiplier=0 does not work

[sstoeckl]

In the demo(demo_max_return) you state that multiplier=0 with a risk objective actually calculates the Standard deviation but does not use it in the objective function. This however is not true, as disabling the objective changes the weights.

[braverock]

Thank you for the report. After carefully tracing through the code in constrained_objective.R:

For portfolio_risk_objective (e.g. StdDev), the contribution to the objective function value is computed on line 202 as out = out + abs(objective$multiplier) * tmp_measure. When multiplier=0, abs(0) * measure = 0, so the contribution is correctly zero.

So why do weights differ? With stochastic optimizers (DEoptim, random portfolios, GenSA, pso), the code evaluates the objective function for every candidate portfolio, even when multiplier=0. This means:

1. Calling StdDev() (or whichever function) shifts RNG state relative to a run with enabled=FALSE where the call is skipped entirely.
2. In iterative solvers like DEoptim, this RNG shift changes the trajectory of the search — different candidate portfolios are explored, yielding a different (but equally valid) local optimum.

This is expected behavior — both variants optimize the same objective; they simply explore the feasible space differently due to RNG state differences. The comment in the demo script that says multiplier=0 will not be 'used in the objective function' is accurate: the value is not added to out. However, the function is still called (so the measure appears in trace output for plotting purposes).

If you need reproducible equivalence, use set.seed() before both runs with the same seed, or use optimize_method="ROI" which is deterministic.

Summary: The code is behaving correctly. The semantic contract of multiplier=0 is: 'compute this measure and include it in trace output, but do not let it influence the objective value.' With stochastic solvers, any additional computation affects RNG state, so exact weight equality with enabled=FALSE is not guaranteed even with the same seed (different number of RNG calls are made). We will add a note to the documentation clarifying this.