IST Vector Postprocessor

01Overview

Pipeline

Solver output is rarely ready to plot as it stands. Stresses are usually stored per element and displacements per node, tensors often arrive as nine separate scalars, and every field carries some discretisation noise. IST Vector therefore handles a result in four stages: it loads the mesh and its fields, optionally recovers smooth nodal fields from the element data, builds the chosen representation (colours, contours, glyphs or streamlines), and finally exports it.

Once a recovery filter is active, the colour bar, the iso-surfaces, the threshold mask and the deformed shape all draw on the recovered field, so smoothed colours are never laid over unsmoothed data. The screen, the PDF, the MP4 and the saved project are produced from a single scene object, which is why every export matches what the viewport shows. PDF surfaces are written as shaded triangles rather than captured as a screenshot, and video frames are recorded at the viewport’s physical pixel size, on HiDPI displays as elsewhere.

03Load

Input formats

Fields may be nodal or per element; scalar, vector or tensor; steady or transient. The format is recognised from the first bytes of the file rather than from its extension, so a misnamed file still reaches the right reader. Files can be opened by drag-and-drop, with Ctrl+O, or from the list of recent files. All of the readers are part of IST Vector itself.

Element types

Every cell is reduced to one of these eight types, and the mid-edge and mid-face nodes of quadratic elements are ignored. VTK strips are unrolled, polygons with five or more vertices are split into triangles, the node order of VTK_VOXEL and VTK_PIXEL cells is corrected, and structured grids become hexa8 elements (quad4 for slabs). General polygons and polyhedra (nsided, nfaced) are read but not drawn, and ghost cells are skipped. A mesh that has nodes but no elements is drawn as a point cloud of round dots, at a diameter of your choice, and the RBF filter can smooth fields over it.

04Your own data

Python writer for EnSight Gold

If your code writes none of the formats above, ist_ensight_gold.py can do the conversion for you. This Python module ships with IST Vector: you give it the nodes, the element connectivity and the fields, and it writes a complete EnSight Gold data set, consisting of a .case file, the .geo geometry and one file per field. EnSight and ParaView read the result as well.

Minimal example

from ist_ensight_gold import write_ensight_gold

write_ensight_gold(
    "out/beam.case",
    nodes=[(0, 0, 0), (1, 0, 0), (1, 1, 0), (0, 1, 0)],
    elements={"quad4": [(0, 1, 2, 3)]},
    node_data={"temperature": [20.0, 35.0, 50.0, 35.0]},
)

This call writes out/beam.case, out/beam.geo and one file per field. Several element blocks, multiple parts, transient steps and binary output are all handled through further keyword arguments of the same function.

Arguments

Because EnSight Gold cannot represent NaN or ±∞, such values are written as 0.0, and whitespace in field and part names is replaced. Invalid input raises an EnSightError that names the field concerned. Running python ist_ensight_gold.py <directory> writes a small transient demo data set with two parts, which makes a convenient first file to open.

06Export

Export formats

07Vector output

Example PDF exports

The Happy Buddha and Armadillo models are © Stanford University Computer Graphics Laboratory (Stanford 3D Scanning Repository). Stanford permits their use in research and in published images with credit, whereas commercial use requires Stanford’s permission. The 2-D pressure field comes from pyvista/vtk-data (Apache-2.0) and originally from the VTK test suite (BSD-3-Clause). All other examples are the author’s own finite-element results and generated surfaces.

09Requirements

Limitations and requirements

• The program requires Windows 10 21H2 or Windows 11 on x64, with a processor that supports x86-64-v2 (SSE 4.2 and POPCNT), and it checks this at startup. There is no macOS or Linux version. Displays from 1280×768 up to 7680×4320 are supported.
• At least 15 GB of RAM is recommended. On machines with less, the program warns at startup and streams large transients from disk.
• IST Vector is a postprocessor and does not solve anything itself. Its input comes from your own solver or from the example data sets installed with it.
• Per-element fields are not contoured directly; a recovery filter must first move them to the nodes.
• Quadratic elements are drawn using their corner nodes only. General polygons and polyhedra (nsided, nfaced) are loaded but not drawn.
• Meshes with more than 50 million nodes are refused at load.
• A transient shows at most 300 steps. Longer series are thinned evenly, with the first and last steps always kept, and a dialog reports the reduction. A .pvd series that does not fit in free memory is thinned further, or refused if not even two steps fit.
• STL and OBJ files carry geometry only, so fields must come from an EnSight or VTK file.

10Engineering

Implementation notes

• Pure Avalonia, no VTK. The interface is built on the stock Avalonia 11.3 packages from NuGet, with themes and controls written for Avalonia. The 3-D view is Avalonia’s own OpenGL control, and the colour bar, labels and axis triad are drawn over it by Avalonia, so no native render window is embedded. The VTK library is neither linked nor called. The readers, iso-surfaces, streamlines, glyphs, software rasteriser and PDF writer are all the project’s own C# code, and it is this that allows the screen, the PDF and the MP4 to be produced from the same scene.
• Headless core. The parsers, recovery methods, scene builder, software rasteriser and PDF writer live in a library that needs no window, no GPU and no third-party package, so the whole pipeline is tested without a display. The vector PDF writer belongs to this library and relies on neither GL2PS nor any other OpenGL feedback library.
• Deterministic output. Depth sorting and shading never depend on timing, so PDFs and video frames are identical byte for byte across runs and thread counts. Regression tests compare them against stored baselines.
• Tests. Every push runs a corpus of EnSight, VTK, OBJ and STL files together with a set of fuzz cases covering truncated, mislabelled and non-finite input. The current build passes 2,878 tests, of which 2,257 exercise the core and 621 the desktop application.
• Playback memory. Interpolation alternates between two scratch buffers and allocates nothing per frame, and coordinates are stored as a structure of arrays.
• Native AOT. The program ships as a single self-contained, trimmed executable. All saved settings are serialised with source-generated JSON and written atomically. OpenGL calls are confined to the UI thread, and GPU caches are rebuilt from generation counters.
• Code signing. Release builds are Authenticode-signed (SHA-256) with an Individual Validation code-signing certificate and carry an RFC 3161 timestamp, with DigiCert, Sectigo and GlobalSign as timestamp servers. The executable, the first-party DLLs and the installer are all signed. Thanks to the timestamp, the signatures remain valid after the certificate expires, and SmartScreen accepts the download without a warning.