Using your own chemistry

Your steel is on a mill certificate, not in the material catalog. You have a ladle analysis, or a grade designation the catalog has never heard of — AISI 4140, S355, St14 — and you want to know what the transformation behavior looks like.

Phases will take that chemistry. What it will not do is pretend the result is calibrated for it. This page explains where that line sits, because understanding it is the difference between using the tool correctly and misreading it.

Giving Phases a composition#

Type the chemistry. You do not need a special syntax.

I have a steel with 0.22 C, 1.25 Mn, 0.25 Cr and 0.003 B — what is it closest to?

Load AISI 4140 chemistry: 0.40 C, 0.85 Mn, 0.25 Si, 0.95 Cr, 0.20 Mo.

If you also know which catalog grade you want as the starting point, say both. Naming the base grade and the change together is one operation:

Load DC04 with C = 0.15 and Mn = 1.2.

That takes DC04's full chemistry and applies your overrides on top. If you leave the base grade out, Phases picks the closest one from your numbers alone, which may not be the one you had in mind.

You only need to give the elements you care about. Everything else comes from the base grade, and iron is balanced to 100% automatically unless you supply Fe yourself. If your numbers add up past 100.5%, the request is rejected: "Composition total is {total}%, which exceeds 100%. Adjust values or omit Fe to auto-balance."

Twenty element rows exist on a material card, and those are the only ones you can set: C, Mn, Si, Ni, Cr, Mo, Cu, S, P, Al, As, W, V, Ti, B (written Bo on the card), Nb, N, O, Fe, Zr. Anything outside that list has nowhere to go.

What the closest-grade match actually is#

Phases does two separate things with your chemistry, and they are easy to confuse.

It names the closest steel. Your composition is screened against a catalog of named grades and the nearest one is reported back — this is identification, so you know roughly what you are holding.

It picks a compatible template. A family classifier runs first and restricts the search: boron at 0.0005 wt% or above routes to press-hardening; carbon at or below 0.12% with Nb, Ti or V at 0.01% or above routes to HSLA; 0.06–0.18% C with Mn at 1.0% or above routes to dual phase; carbon at or below 0.10% with Mn at or below 0.5% routes to mild. Family first, then nearest neighbor within it — which is why a boron steel is not matched to a non-boron neighbor that happens to sit closer on carbon and manganese. Ambiguous chemistry falls back to a search across the whole catalog.

Neither of those is a physics claim. The product's own instruction is blunt about it: "This identity/template match is not a physics-calibration claim."

You will also see a template-match badge — good, moderate, or high deviation, with a numeric distance. Read it as one thing only: how far your chemistry sits from the nearest catalog template. It is explicitly not an accuracy score. In the product's words, "It is not a kinetics-accuracy or calibration score; simulator capability results are authoritative."

Where the deviation is large you get told so in plain text, for example: "Significant deviation from closest grade {GRADE} (distance={d}). Template similarity is low; physics availability is evaluated separately when requested." Chemistry that has wandered out of the catalog's territory gets its own warnings — above 5% Cr, above 5% Ni, or above 0.6% C each add a note that you are heading toward stainless, austenitic or tool-steel country.

The confirmation gate#

Any custom or changed chemistry stops before anything is created. The tool returns a needs_confirmation status and builds nothing. This is deliberate and it is not skippable — passing a custom composition straight into a simulation to route around it is forbidden by the agent's instructions.

The message you get for a new chemistry is fixed:

This chemistry is closest to {grade}. Its stable material definition and compatible base-card template can be saved now; each simulation will resolve only the physics it needs and will report unavailable or fallback states explicitly. Continue or adjust the composition?

If you edit an already-active material instead, the wording is:

This modified chemistry is closest to {grade}. Saving the definition will not calculate physics; the next simulation will request and report only its required material capabilities. Continue or adjust?

Alongside the message you get the identified grade and family, the template family, the top template candidates, the distance to the nearest grade, your normalized composition — and three fields that say the quiet part out loud: calibrated: false, physics_status: "not_requested", and an empty capabilities set. If no named steel matches at all, the grade is reported as an unrecognized chemistry.

What confirming creates — and what it does not#

The instruction the agent follows says it directly: "Confirming creates a stable definition and base template only; it does not calculate kinetics, phase flow stress, or V-Gleeble surfaces."

How to confirm#

Say yes. "Continue", "go ahead", "yes, create it" — the agent re-runs the same call with confirmation set, and that is the whole handshake. One edit costs at most two steps.

Often you get a Custom material card instead of a plain question: a Name field, a read-only base grade line, and an editable Composition (wt %) grid covering C, Mn, Si, Cr, Mo, Ni, Bo, Al, V, Ti and Nb plus any extra elements the agent filled in, with Fe shown as the computed balance. Adjust the numbers there if you want, then press Activate for this session. Its other button, Save to workspace for future use, persists the recipe to your workspace library — see workspace materials.

If you decline#

Nothing is created. You are offered two routes: adjust the chemistry, or pick a different grade. No material is stored and no simulation is run.

After you confirm#

The confirmed material becomes the active session material. Every simulation tool — heat treatment, CCT, RSW, V-Gleeble — uses it automatically when you do not name a grade explicitly. The chat shows Active material: {name} directly above the message composer so you always know what is loaded. You do not need to re-load or re-list it between steps.

One thing confirming does not do is start a simulation. Creating, loading, modifying, comparing or saving a material is never read as a request to run anything; the agent stops and asks which simulator you want. See how the conversation works.

To change it later, edit it in place:

Change the carbon on my active material to 0.25%.

Chemistry edits go through the same gate. Initial-phase edits do not — but the five phase fractions (Ferrite, Pearlite, Bainite, Martensite, Austenite) must sum to 1.0 within 0.05, or you get "Phase fractions sum to {x}, expected 1.0. Adjust values so they balance."

If the material you are editing came from your workspace library, the edit forks it into a session-only copy. The saved workspace row is never quietly rewritten by a chat edit — persisting the fork is a separate, explicit save.

When a simulator says a capability is unavailable#

This is the part that surprises people. Confirming was cheap because it resolved no physics. Each simulator resolves its own requirements at run time, and it will tell you honestly when it cannot.

Full detail lives in the capability matrix. Three outcomes are worth knowing in advance:

Kinetics fall back to the template's own. For heat treatment, CCT, V-Gleeble thermal modes and microstructure-enabled RSW, if no published kinetics model applies to your chemistry, the run keeps the verified template's authored kinetics and reports that as a warning-carrying preserved_source result. That is source preservation, stated explicitly. It is not interpolation to your chemistry, and it is not calibration.

V-Gleeble deformation modes refuse custom chemistry outright. Hot tensile, hot compression and Satoh need a validated bulk flow surface — an exact per-grade table, or an explicitly reviewed grade mapping. A custom chemistry gets: "No validated custom V-Gleeble bulk flow surface exists for {material}; phase P1-P4 is not a compatible substitute." A calculated phase flow stress cannot stand in for it — they are different capabilities. Thermal modes still work. See virtual Gleeble.

RSW needs a supported base grade underneath. RSW's electrical, thermal and mechanical property database covers a fixed list of grades. Your custom chemistry rides on one of them: the plate's grade stays a supported grade such as DP600, and the chemistry travels alongside it for the microstructure calculation. A material with no validated RSW bulk-property mapping is rejected — "This material has no validated RSW bulk-property mapping." See resistance spot welding.

There is also a hard stop earlier in the chain. If no verified template in the catalog can represent every non-zero element you gave, nothing is generated: "No verified template can represent the requested composition." followed by why each candidate failed. Removing exotic elements, or moving to a chemistry inside a catalog family, is the way through.

When any of these come back, the practical next steps are in common problems. The boundaries of what the solvers model at all are in known limits — worth reading once before you rely on a custom-chemistry result.

The short version#

Phases will run your chemistry. It names the closest steel, mounts your composition on a verified template, and tells you how far apart they are. It will not tell you that makes the prediction calibrated for your steel, because it does not. Every simulator states separately what it resolved, what it substituted, and what it could not do — and that statement, not the match badge, is the thing to read.