Parts of the adjoint optimisation library were re-designed to generalise
the way sensitivity derivatives (SDs) are computed and to allow easier
extension to primal problems other than the ones governed by
incompressible flows. In specific:
- the adjoint solver now holds virtual functions returning the part of
SDs that depends only on the primal and the adjoint fields.
- a new class named designVariables was introduced which, apart from
defining the design variables of the optimisation problem and
providing hooks for updating them in an optimisation loop, provides
the part of the SDs that affects directly the flow residuals (e.g.
geometric variations in shape optimisation, derivatives of source
terms in topology optimisation, etc). The final assembly of the SDs
happens here, with the updated sensitivity class acting as an
intermediate.
With the new structure, when the primal problem changes (for instance,
passive scalars are included), the same design variables and sensitivity
classes can be re-used for all physics, with additional contributions to
the SDs being limited (and contained) to the new adjoint solver to be
implemented. The old code structure would require new SD classes for
each additional primal problem.
As a side-effect, setting up a case has arguably become a bit easier and
more intuitive.
Additional changes include:
---------------------------
- Changes in the formulation and computation of shape sensitivity derivatives
using the E-SI approach. The latter is now derived directly from the
FI approach, with proper discretization for the terms and boundary
conditions that emerge from applying the Gauss divergence theorem used
to transition from FI to E-SI. When E-SI and FI are based on the same
Laplace grid displacement model, they are now numerically equivalent
(the previous formulation proved the theoretical equivalence of the
two approaches but numerical results could differ, depending on the
case).
- Sensitivity maps at faces are now computed based (and are deriving
from) sensitivity maps at points, with a constistent point-to-face
interpolation (requires the differentiation of volPointInterpolation).
- The objective class now allocates only the member pointers that
correspond to the non-zero derivatives of the objective w.r.t. the
flow and geometric quantities, leading to a reduced memory footprint.
Additionally, contributions from volume-based objectives to the
adjoint equations have been re-worked, removing the need for
objectiveManager to be virtual.
- In constrained optimisation, an adjoint solver needs to be present for
each constraint function. For geometric constraints though, no adjoint
equations need to solved. This is now accounted for through the null
adjoint solver and the geometric objectives which do not allocate
adjoint fields for this kind of constraints, reducing memory
requirements and file clutter.
- Refactoring of the updateMethod to collaborate with the new
designVariables. Additionally, all updateMethods can now read and
write restart data in binary, facilitating exact continuation.
Furthermore, code shared by various quasi-Newton methods (BFGS, DBFGS,
LBFGS, SR1) has been organised in the namesake class. Over and above,
an SQP variant capable of tackling inequality constraints has been
added (ISQP, with I indicating that the QP problem in the presence of
inequality constraints is solved through an interior point method).
Inequality constraints can be one-sided (constraint < upper-value)
or double-sided (lower-value < constraint < upper-value).
- Bounds can now be defined for the design variables.
For volumetricBSplines in specific, these can be computed as the
mid-points of the control points and their neighbouring ones. This
usually leads to better-defined optimisation problems and reduces the
chances of an invalid mesh during optimisation.
- Convergence criteria can now be defined for the optimisation loop
which will stop if the relative objective function reduction over
the last objective value is lower than a given threshold and
constraints are satisfied within a give tolerance. If no criteria are
defined, the optimisation will run for the max. given number of cycles
provided in controlDict.
- Added a new grid displacement method based on the p-Laplacian
equation, which seems to outperform other PDE-based approaches.
TUT: updated the shape optimisation tutorials and added a new one
showcasing the use of double-sided constraints, ISQP, applying
no-overlapping constraints to volumetric B-Splines control points
and defining convergence criteria for the optimisation loop.
- the faMesh/fvMesh copy constructors were using the readOption from
the base-mesh schemes/solution instead of copying their contents.
This would not really affect fvMesh (since it has its own IOobject
for the constructor), but did affect faMesh. However, the problem
only shows up with collated + redistribute, since that is where
the ranks can be doing uncoordinated IO.
Only consider as a bug for recent develop since previous versions
had other problems with collated+redistribute with finite-area
anyhow.
- retain group information when copying zones
- support construct empty (add details later)
- improve consistency for zone and boundaryMesh construction
- support front/back/both selection for faceZoneToCell
STYLE: prefer faceZone patch() method instead of operator()
STYLE: use std::unique_ptr instead of manual pointer management
- for zones and core patch types.
Easier data management, allows default destructors (for example)
- update TimeState access methods
- use writeTime() instead of old method name outputTime()
- use deltaTValue() instead of deltaT().value()
to avoids pointless construct of intermediate
- for simulations where the yPlus is needed for other purposes or
just for obtaining information on the patches it can be useful
to disable field writing and save disk space.
The 'writeFields' flag (as per some other function objects)
has been added control writing the yPlus volume field.
If unspecified, the default value is 'true' so that the yPlus
function object continues to work as before.
However, this default may change to 'false' in the future to align
with other function objects.
ENH: wallShearStress: support disable of field writing
- similar to yPlus, the write() method combines writing information
and writing the fields. The 'writeFields' flag allows some
separation of that logic.
- enable 'faceZones' support.
- enable 'writeFile' support to better control file output.
- rename 'PatchPostProcessing' as 'ParticlePostProcessing' for better clarity.
- fix#1808
- enable 'faceZone' support.
- introduce 'cloudFunctionObjectTools' to simplify collection of particle info
on patches or face zones.
- enable 'writeFile' support to better control file output.
- rename 'PatchParticleHistogram' as 'ParticleHistogram' for better clarity.
- accept plain lists (space or comma separated) as well as the
traditional OpenFOAM lists. This simplifies argument handling
with job scripts.
For example,
simpleFoam -ioRanks 0,4,8 ...
vs
simpleFoam -ioRanks '(0 4 8)' ...
It is also possible to select the IO ranks on a per-host basis:
simpleFoam -ioRanks host ...
- expose rank/subrank handling as static fileOperation methods
- sometimes the last commit is not enough information about
the tested state (especially with extensive rebasing).
Also provide the short context of some previous commits.
- provide uniformMixed conditions for finite-area and finite-volume.
These are intended to replace the exprMixed condition but allow
the full range of different PatchFunction1 and Function1 types.
- add uniformFixedGradient to finite-area for completeness.
Note:
- still some possible difficulties with the order of evaluation.
- eg, using an expression within the 'U' field that depends
of the surface 'phi' field before that is constructed.
In this case, the 'value' entry is really needed.
- the default (uninitialised) value for edge connections of -1
could be confused with a tagged finiteArea patch, which used
(-patchid-1) encoding. This would lead to messages about erroneous
processor-processor addressing, but is in fact an mismatched edge
connection.
Now tag the finiteArea patch as (-patchid-2) to avoid this ambiguity
and correctly generate an "Undefined connection:" message instead.
Properly flush the VTP writers before raising a FatalError
to ensure that they are not prematurely truncated.
Open Point:
The base problem of "Undefined connection:" is largely related to
multiply-connected face edges (ie, from the underlying volume mesh).
Not easily remedied in the finiteArea generation.
TUT: basic finiteArea setup on motorBike
- with (nPollProcInterfaces < 0) it does the following:
- loop, waiting for some requests to finish
- for each out-of-date interface, check if its associated
requests have now finished (ie, the ready() check).
- if ready() -> call updateInterfaceMatrix()
In contrast to (nPollProcInterfaces > 0) which loops a specified
number of times with several calls to MPI_Test each time, the
(nPollProcInterfaces < 0) variant relies on internal MPI looping
within MPI_Waitsome to progress communication.
The actual dispatch still remains non-deterministic (ie, waiting for
some requests to finish does not mean that any particular interface
is eligible for update, or in any particular order). However, using
Waitsome places the tight looping into the MPI layer, which results
in few calls and eliminates behaviour dependent on the value of
nPollProcInterfaces.
TUT: add polling to windAroundBuildings case (for testing purposes)
- separate broadcast times from reduce/gather/scatter time
- separate wait times from all-to-all time
- support invocation counts, split off requests time/count
from others to avoid flooding the counts
- support 'detail' switch to increase the output information.
Format may change in the future
