Wiring diagrams


Data structure for (directed) wiring diagrams, aka string diagrams.

A (directed) wiring diagram consists of a collection of boxes with input and output ports connected by wires. A box can be atomic (possessing no internal structure) or can itself be a wiring diagram. Thus, wiring diagrams can be nested recursively. Wiring diagrams are closely related to what the CS literature calls "directed graphs with ports" or more simply "port graphs". The main difference is that a wiring diagram has an "outer box": a wiring diagram has its own ports that can be connected to the ports of its boxes.

This module provides a generic data structure for wiring diagrams. Arbitrary data can be attached to the boxes, ports, and wires of a wiring diagram. The diagrams are "abstract" in the sense that they cannot be directly rendered as raster or vector graphics. However, they form a useful intermediate representation that can be serialized to and from GraphML or translated into Graphviz or other declarative diagram languages.


Grapn underlying wiring diagram, including parts for noin-internal wires.

The graph has two special vertices representing the input and output boundaries of the outer box.


Create an encapsulating box for a set of boxes in a wiring diagram.

To a first approximation, the union of input ports of the given boxes will become the inputs ports of the encapsulating box and likewise for the output ports. However, when copies or merges occur, as in a cartesian or cocartesian category, a simplification procedure may reduce the number of ports on the encapsulating box.


  1. Each input port of an encapsulated box will have at most one incoming wire

from the encapsulating outer box, and each output port of an encapsulated box will have at most one outgoing wire to the encapsulating outer box.

  1. A set of ports connected to the same outside (non-encapsulated) ports will be

simplified into a single port of the encapsulating box.

See also induced_subdiagram.


Operadic composition of wiring diagrams.

This generic function has two different signatures, corresponding to the "full" and "partial" notions of operadic composition (Yau, 2018, Operads of Wiring Diagrams, Definitions 2.3 and 2.10).

This operation is a simple wrapper around substitute.


Wire together two ports in an undirected wiring diagram.

A convenience method that creates and sets junctions as needed. Ports are only allowed to have one junction, so if both ports already have junctions, then the second port is assigned the junction of the first. The handling of the two arguments is otherwise symmetric.

FIXME: When both ports already have junctions, the two junctions should be merged. To do this, we must implement merge_junctions! and thus also rem_part!.


Wiring diagrams as a symmetric monoidal category.

This module provides a high-level categorical interface to wiring diagrams, building on the low-level imperative interface and the operadic interface. It also defines data types and functions to represent diagonals, codiagonals, duals, caps, cups, daggers, and other gadgets in wiring diagrams.


Crossing minimization by sorting a univariate statistic.

The boxes in sources and/or targets are fixed and the boxes in vs are permuted. A permutation σ of the latter is returned, such that vs[σ] are the sorted box IDs. Both one-sided and two-sided crossing minimization are supported, depending on whether just one, or both, of sources and targets are given.

In this simple but popular heuristic algorithm, the boxes are permuted by sorting a univariate statistic of the positions of incoming and/or outgoing wires. Typical choices are:

  • mean: the sample mean, yielding the "barycenter method"
  • median: the sample median

In both cases, this algorithm has the property that if there is a permutation with no crossings, it will find it.


Normalize copies in a wiring diagram.

This function maximizes sharing of intermediate computations in a wiring diagram where copies are natural.

This algorithm is basically the same as the congruence closure algorithm on term graphs, in the special case of the empty relation R = ∅ (Baader & Nipkow, 1998, Term Rewriting and All That, Sec. 4.4). The main difference is the possibility of zero or many function outputs.


Conventions for serialization of wiring diagrams.

Defines a consistent set of names for boxes, ports, and wires to be used when serializing wiring diagrams, as well as conventions for serializing box, port, and wire attributes.


Serialize abstract wiring diagrams as GraphML.

Serialization of box, port, and wire values can be overloaded by data type (see convert_to_graphml_data and convert_from_graphml_data).

GraphML is the closest thing to a de jure and de facto standard in the space of graph data formats, supported by a variety of graph applications and libraries. We depart mildly from the GraphML spec by allowing JSON data attributes for GraphML nodes, ports, and edges.


  • GraphML Primer: http://graphml.graphdrawing.org/primer/graphml-primer.html
  • GraphML DTD: http://graphml.graphdrawing.org/specification/dtd.html

Serialize abstract wiring diagrams as JSON.

JSON data formats are convenient when programming for the web. Unfortunately, no standard for JSON graph formats has gained any kind of widespread adoption. We adopt a format compatible with that used by the KEILER project and its successor ELK (Eclipse Layout Kernel). This format is roughly feature compatible with GraphML, supporting nested graphs and ports. It also supports layout information like node position and size.


  • KEILER's JSON graph format: https://rtsys.informatik.uni-kiel.de/confluence/display/KIELER/JSON+Graph+Format
  • ELK's JSON graph format: https://www.eclipse.org/elk/documentation/tooldevelopers/graphdatastructure/jsonformat.html