Tutorial 4: A sparse matrix from SuiteSparse

A gallery of six SuiteSparse matrices laid out with Yifan Hu (SFDP) in 3D and colored by eigenvector centrality.

The SuiteSparse Matrix Collection holds over 2,800 sparse matrices from real-world applications: finite-element meshes, circuit simulations, optimization problems, and others.

1. Import a matrix

  1. Open SciGraphs ▸ Data and set the source to SuiteSparse.
  2. Enter a Matrix Identifier in Group/Name form (e.g. Grund/bayer09). Use Browse Collection to find identifiers.
  3. Choose the Graph Representation:
    • Symmetric interprets the matrix as a standard adjacency graph via \(A + A^\top\) (denser, rounder layouts).
    • Bipartite treats rows and columns as two disjoint node sets, preserving the original matrix structure (elongated layouts).
  4. Enable Giant component only to keep just the largest connected component.
  5. Optionally enable Auto layout on import and pick Yifan Hu as the default layout.
  6. Press Download & Create Graph.
Note

If the archive ships an auxiliary coordinate file, SciGraphs applies those coordinates directly as vertex positions and skips the layout entirely, reproducing the geometry of the underlying physical domain.

2. Lay out (matrices without coordinates)

For matrices without coordinate data, open Layout & Positioning and apply Yifan Hu (sfdp) in 3D, which typically produces the most informative results for large sparse structures.

3. Color by eigenvector centrality

  1. Run Setup Visual (Data panel, Step 4) if you have not already.
  2. In Analysis ▸ Centrality Metrics, choose eigenvector and press Calculate.
  3. In the Visualization panel, map node color to the eigenvector-centrality attribute using a colormap such as Inferno or Black-Body Radiation. Bright nodes mark the vertices most central to the dominant eigenstructure; dark regions are peripheral.

4. Render

Add lighting and a camera, then render with Cycles. For dense meshes, depth of field and adaptive text labels (label only high-degree nodes via the Attribute Filter) reduce clutter and reinforce depth perception.

For a comparative figure like the one above, repeat the import for several identifiers from different problem domains and arrange the resulting objects in one scene.

Back to top