Process Systems Enterprise Limited
email this page print this page
pdf overview

gPROMS v3.1: Execution and solution features

Faster, easier and more robust

Execution
         & solution

On-site workshop?
 Click to arrange

Update via WebEx?
 Click to arrange

As well as enhanced flowsheeting, display and modelling features, gPROMS v3.1 embodies a number of significant enhancement on the solution front – including unified solver and kernel architecture, the ability to solve high-index problems and improvements to handling of certain equation formulations.

This page outlines some of the key features. For more information see the release notes or request an on-site update for your company.

Features at-a-glance

Details

Click on the images below to enlarge

 

 

 

 

 

Automatic index reduction for high-index problems – analysis and notification

 

 

 

 

 

 

Execution output – three different views – with easy querying of variable and unit information

 

Automatic index reduction

The newly-introduced automatic index reduction affects models of so-called high index. High-index problems typically have fewer degrees of freedom with respect to specifying initial conditions than 'normal' DAE problems and can not be solved directly.

Up to version 3.0 gPROMS would report a high-index problem and stop execution. The high index could then be manually removed from the model.

The new index reduction feature will now automatically address and resolve the index problem.

gPROMS's automatic index reduction works in two stages:

All the changes are reported in the diagnostic execution output window.

Combined execution output view

The execution output window is now a combined view of

This combined view has been introduced to give quick access to all key results by selecting different tabs of the same window.

Easy querying of variables and units

It is now possible to use the variable name rather than the variable numbers for queries, by entering the fully-qualified path – e.g. 'Flowsheet.heater.temperature'.

By pressing 'CTRL+SPACE' when entering variable names, gPROMS will either list the available variables (or units) or complete the path name if only one choice is available.

Similar assisted pathname completion applies when querying entire units.

Enhanced execution output context menu

The context menu in the execution output has also been enhanced.

If a variable name is displayed in the execution output the context menu now offers the additional option to query the variable or a unit selected by clicking on the variable or unit name.

Unified solver architecture

gPROMS v3.1 introduces a single set of solvers for all activities: simulation, optimisation, parameter estimation and optimal design of experiments.

The immediate benefit of this change are consistent behaviour and diagnostics throughout, plus consistent application of the various other enhancements listed below across all gPROMS activities.

The new solver architecture is based on the same (new) nonlinear block decomposition solver BDNLSOL and the nonlinear (sub-)solver SPARSE. Although the solver NLSOL is no longer available, backward compatibility is ensured by:

This can be controlled by solution parameters which are turned on by default.

Conservative discretisations of PDEs

In the discretisation approach prior to gPROMS v3.1, complex terms were automatically expanded before a discretisations was performed, as follows:

This approach had the disadvantage of not being a non-conservative formulation which may have led to errors in conservation equations for total mass, energy, momentum etc.

From v3.1 onwards, gPROMS applies discretisations on the entire term which leads to a conservative formulation without the necessity to increase memory requirements through the introduction of intermediate variables:

Changes to indexing of continuous domains

Under certain circumstances a continuous index could be used to index a discrete array. Consider the following example: 

  FOR x := 0 TO test_max DO
     test(x) = x ; # (OK)
     test1(x) = test(index(x)); # (INCORRECT)
  END

Here 'x' is a continuous index when used in 'test(x)' which is the correct usage. 'Index', however, is a parameter array and therefore a discrete domain. From v3.1 onwards this is not permitted.

New convergence criterion for Optimisation, Parameter Estimation and Optimal Design of Experiments

The non-linear optimisation solver SRQPD used by Optimisation, Parameter Estimation and Optimal Design of Experiments has been enhanced with a new convergence criterion.

The criterion used to date (which is still available and the default) is based on an estimation of the improvement whereas the new criterion is based on a combination of an optimality tolerance and a tolerance for the improvement between iterations.

Depending on the optimisation problem, the new criterion may lead to better results.

The corresponding solution parameter of SRQPD are: 

  "ConvergenceCriterion" := ImprovementEstimateBased; 
  "ConvergenceCriterion" := OptimalityBased; (new)

For the new criterion a new parameter for specifying the improvement tolerance has been introduced

  "NoImprovementTolerance" := 1e-012;

Handling of continuous switches in discontinuities

A new solution parameter for the Sparse solver has been introduced to limit "chattering" in conditional equations (If/Then/Else or Case statements). This controls the maximum number of switches: 

    DASolver := "DASOLV" [
        "InitialisationNLSolver" := "BDNLSOL" [
            "BlockSolver" := "SPARSE" [
                "MaxStructureSwitches" := 100;
            ]
        ],
        "ReinitialisationNLSolver" := "BDNLSOL" [
            "BlockSolver" := "SPARSE" [
                "MaxStructureSwitches" := 100
            ]
        ]
    ]

 Request a WebEx or on-site update