More precisely, incorrect equation normalization could occur in the presence of
cos, sin, tan, cosh and x^n (where n is an even integer).
Also add some comments explaining why some other rules are (hopefully) correct.
The new representation is only supported for MATLAB/Octave, C and Julia output
for the time being. Bytecode and JSON are unsupported.
This commit adds new fields in M_.
This is a preliminary step for dynare#1859.
Should have no impact though, since diff nodes are already substituted out at
that point. But it’s better to implement it properly, in case we change the
substitution rules later.
By the way, make the computeSubExprContainingVariable method protected.
– factorize common code between the static and the dynamic version
– reorganise language-specific code into dedicated functions
– use a function template in the main helper to do some computations
at compile-time (using constexpr features)
When the same complex expression appears outside and inside a steady_state()
operator, the same temporary term would be used for both cases, which was
obviously wrong.
The fix consists in never substituting temporary terms for expressions inside
the steady_state operator().
Incidentally, this implies that external functions can no longer be used inside
steady_state operators (since their computed values are stored inside temporary
terms).
– a generic one: CommonEnums.hh
– and a bytecode-specific one: Bytecode.hh
By the way, rename global constant “near_zero” into “power_deriv_near_zero”,
for clarity.
In particular, use this feature in many loops which feature a special treatment
for the first iteration, using a boolean variable (replacing iterator
manipulation). By the way, also use std::exchange() to simultaneously test the
value of this variable and update it.
– also use vectorized versions for .+ and .-
– add a space before vectorized operators to avoid syntactical ambiguity
This commits complements bfdcc546ec.
When an endogenous is declared with “var(log)”, say “y”:
– creates an auxiliary named “LOG_y”
– replaces “y(±l)” everywhere by “exp(LOG_y(±l))”
– adds a new auxiliary equation “y=exp(LOG_y)”
– adds a new definition “LOG_y=log(y)” in set_auxiliary_variables.m and
dynamic_set_auxiliary_series.m files
This option also works in conjunction with “deflator=…”, such as “var(log,
deflator=…)” (the “log” must appear befor “deflator”). There are no provisions
for combining “log” with “log_deflator”, because that would not make much sense
from an economic point of view (amounts to taking the log two times).
Ref. dynare#349
This is a more natural semantics.
Incidentally, this fixes a bug in the variable mapping (M_.mapping) where some
endogenous, appearing in a log() in a VAR or TCM, would not be mentioned (e.g.
in the var-expectations/7/example1.mod test, and many others).
By the way:
– Fix and improve the explanation of the purpose of the orig_symb_id and
orig_lead_lag fields for auxvars
– Factorize the code that prints those fields in MATLAB and JSON output
The logic was flawed in several ways. In particular, the test files
pac/trend-component-{3,10,11}/example1.mod would return A0 and A0star matrices
where the (2,2) element was incorrectly zero.
Sébastien Villemot