Calculation workflows

#Open the calculation canvas

Start with Build a silicon supercell with AI to author and run a graph through conversation, or upload a structure and build a supercell to practice manual configuration.

Open Research Apps → Material Expert. Select an existing workflow in the sidebar or create a new one. You can also start with a template.

Give the workflow a name that identifies its system and purpose, such as “Si · Geometry optimization.” Check the save information above the canvas after editing, before leaving or running it.

#Build through an AI conversation

Open the calculation assistant on the right and describe the system, available structure, steps and outputs. It can read the current graph, create or change nodes and connections, and check or run the workflow when requested.

Ask it to build and check without running while deciding on configuration. Explicitly request execution when ready. Inspect the resulting settings manually on the same canvas. See Use the AI calculation assistant for requests, conversations and results.

#Configure nodes and connections

Select Add node to browse the node library. Categories include structure input and preparation, DFT calculations, molecular dynamics, analysis and output. Read each node's description and availability; input and execution requirements differ.

Select a node on the canvas, configure its parameters and check that upstream outputs supply what the next node needs. Pay particular attention to:

• Elements, cell, atom count and periodic boundaries.
• Units and the applicability of the method, force field or model.
• Required parameters and file references.
• Connections between input, processing and output nodes.

Nodes marked as awaiting execution support, such as the active-learning loop, cannot yet be run.

See Canvas editing for controls, Workflow nodes for categories, and Inputs, variables and outputs for field configuration and references.

Chinese interface. Select a node to inspect its upstream input and parameters on the right; validation and test-run controls are above the canvas.

#Validate and test

1. Select Check in the toolbar and resolve missing inputs or connection issues.
2. Confirm that the draft is saved, then select Test run.
3. Enter the values requested by the user-input node in the run-input panel.
4. Check the inputs and calculation size before starting the run.

If prompted to add a user-input node, return to the canvas and connect one. A test run executes the saved draft; choose a calculation size appropriate for what you are checking.

Issues in the check panel can locate the relevant node. Recheck after changing parameters and run the latest saved draft. See Checks, runs and node traces for single-step verification, inputs and failed nodes.

#Inspect a run

Open Run history and select an execution. The test panel's results, run details and node trace show outputs, execution information and individual node states.

Open the failed node's logs. Supply missing inputs before running again; follow the logged cause for other errors. If an output is absent, first check whether its node ran.

A completed run means execution finished. Before drawing a research conclusion, also check the structure, convergence, parameter sensitivity and output units. See Read and check research results.

#Preserve a release

To keep a version of the configuration, select Publish, add release notes and submit. Review or restore it later in the version history.

Restoring an older version affects the current draft. Preserve any changes you need before restoring. A release does not validate the scientific result. To reuse a configuration as a template, add it separately in the Template Center.

See Versions and reuse for the differences between saving, publishing, restoring and adding templates.