Jacobi Docs — Feature Reference & Examples

Overview

What is Jacobi?

Jacobi is a desktop IDE purpose-built for engineers who write Abaqus user subroutines. It replaces the manual edit → compile → submit → wait → inspect loop with an instant single-point RVE test that tells you in seconds whether your constitutive model is physically correct — and exactly which line of Fortran is wrong if it isn't.

⚡

Instant physics tests

Run 35 single-point RVE tests against your subroutine in under 3 seconds, without touching Abaqus.

🔬

Subroutine-aware editor

Monaco editor with Fortran syntax, static analysis, live squiggles, and Abaqus argument documentation on hover.

🤖

AI diagnostics

Claude Sonnet reads your test results and points to the exact Fortran variable that's wrong, with a suggested fix citing the canonical textbook.

📐

5 subroutine types

Fully auto-detected. UMAT, VUMAT, UEL, UHARD, and CREEP each get their own test suite and diagnostic logic.

Quickstart

Get running in 60 seconds

Jacobi requires gfortran on your PATH. Everything else is self-contained.

1

Open your subroutine file

Use Ctrl+O or File → Open. Jacobi accepts .f, .f90, .for, and .f95. It immediately detects which Abaqus subroutine type the file contains.

2

Check the Subroutine Panel

The right sidebar shows the detected subroutine (UMAT, VUMAT, etc.) with all arguments listed. Any interface violations are flagged immediately — wrong number of arguments, wrong types, missing IMPLICIT NONE.

3

Open the Tests tab → Run Tests

Click the Tests tab in the bottom panel. Set material parameters (E, ν, σ_y) and click Run Tests. Jacobi compiles your subroutine with gfortran, runs a single-point RVE driver, and parses the results in one shot.

4

Read the results

Each test card shows PASS or FAIL with the measured vs expected value, a chart, the exact lines from your code being tested, and a Testing: banner describing the physics being checked.

5

Click "Analyze with AI" on any failure

Claude Sonnet receives only the failing tests, with measured values and canonical references. It responds with the exact Fortran variable that is wrong and a 4–6 line fix in Fortran 90 syntax.

ℹ

Prerequisites: gfortran 9+ on your system PATH. On Windows, install via MSYS2. On macOS: brew install gcc. On Linux: apt install gfortran.

Editor

Fortran Editor

Jacobi uses Monaco (the VS Code engine) with a custom Fortran grammar, Abaqus-aware hover documentation, and live diagnostics.

Syntax highlighting

Full tokenization of Fortran 77, 90, and 95 — keywords, intrinsics, strings, comments, preprocessor directives, and numeric literals including REAL*8, D-notation, and complex literals.

      Fortran 90
      umat_plasticity.f90
    

!  J2 plasticity UMAT — return-mapping algorithm
SUBROUTINE UMAT(STRESS, STATEV, DDSDDE, SSE, SPD, SCD,
     &           RPL, DDSDDT, DRPLDE, DRPLDT,
     &           STRAN, DSTRAN, TIME, DTIME, TEMP, DTEMP,
     &           PREDEF, DPRED, CMNAME, NDI, NSHR, NTENS,
     &           NSTATV, PROPS, NPROPS, COORDS, DROT,
     &           PNEWDT, CELENT, DFGRD0, DFGRD1,
     &           NOEL, NPT, LAYER, KSPT, KSTEP, KINC)

  IMPLICIT NONE
  INCLUDE 'ABA_PARAM.INC'

  REAL*8 STRESS(NTENS), STATEV(NSTATV), DDSDDE(NTENS,NTENS)
  REAL*8 PROPS(NPROPS), DSTRAN(NTENS)

  !  Material constants from PROPS array
  REAL*8 EMOD, ANU, SYIELD0, HMOD
  EMOD   = PROPS(1)   !  Young's modulus (MPa)
  ANU    = PROPS(2)   !  Poisson's ratio
  SYIELD0 = PROPS(3)  !  Initial yield stress (MPa)
  HMOD   = PROPS(4)   !  Linear hardening modulus (MPa)

Static analysis

Jacobi runs a full static pass on every keystroke. Diagnostics appear as Monaco squiggles and in the Problems panel.

Diagnostic	Severity	Example trigger
Undefined variable used before assignment	error	`X = Y + 1` where `Y` is never assigned
Variable declared but never used	warn	`REAL*8 TEMP` declared, never referenced
Possible typo — variable not in scope	error	`EMDOD` instead of `EMOD`
Missing IMPLICIT NONE	warn	No `IMPLICIT NONE` at top of subroutine
UMAT/VUMAT interface argument count mismatch	error	Wrong number of arguments in SUBROUTINE line
Variable assigned but value never read	warn	`DLAM = 0.0` overwritten before use

✦

Click any squiggle to jump to the problem, or press F8 to cycle through all diagnostics. The Problems count in the status bar updates live.

Editor

Themes

20 editor themes — 18 dark, 2 light. Switch instantly via the command palette (Ctrl+Shift+P) and the selection persists across sessions.

Dark+

Monokai

Dracula

One Dark Pro

Nord

GitHub Dark

Tokyo Night

Catppuccin

Rosé Pine

Poimandres

Kanagawa

Night Owl

Switch via Ctrl+Shift+P → type "theme" → select any option. The setting persists across sessions.

Editor

Command Palette

Every action in Jacobi is accessible via the command palette. Press Ctrl+Shift+P anywhere.

Command palette categories

File
  Open File…                   Ctrl+O
  Save File                    Ctrl+S

Build
  Build (compile + run)        F5

View
  Zoom In / Out / Reset        Ctrl++ / Ctrl+- / Ctrl+0

Theme  (20 entries, one per theme)
  Dark+, Monokai, Dracula, One Dark Pro, Nord, GitHub Dark …

Editor

Subroutine Templates

Generate a correctly-structured Fortran scaffold for any Abaqus subroutine type. All arguments, array dimensions, and common patterns pre-filled.

1

Open the Template Dialog

Click the New from Template button in the toolbar, or use the command palette → "New Template".

2

Select subroutine type

Choose UMAT, VUMAT, UEL, UHARD, CREEP, or any other supported Abaqus subroutine. The dialog shows the argument list and description from the Abaqus manual.

3

Insert into editor

The template opens as a new tab with IMPLICIT NONE, all argument declarations typed correctly, and TODO comments at the key implementation sites.

✦

Templates use free-form Fortran 90 (& continuation) regardless of file extension. If you need fixed-form, save with a .f extension and reformat.

Build

Build System

One-key compilation with gfortran. Error messages are parsed and displayed as Monaco squiggles with click-to-jump navigation.

Running a build

Press F5 or click Build in the toolbar. The Build Console opens in the bottom panel showing the raw gfortran output.

Build console output

▶ Building umat_plasticity.f90…
gfortran -O2 -Wall -fcheck=all -o /tmp/jacobi_out umat_plasticity.f90

✓  Compiled successfully  (0.42s)

─── or, on error ───

umat_plasticity.f90:47:18: error: 'DLMA' is not a member of
  DLMA = DLMA + DF / DENOM
                  ^
→ Click to jump to line 47

Error navigation

Every error line in the Build Console is clickable — clicking it moves the cursor to the exact file and line in the editor. The squiggle appears on the offending token, not just the line.

Build

Integrated Terminal

A full PowerShell (Windows) or bash (macOS/Linux) terminal embedded in the bottom panel. Runs in the directory of the active file.

Terminal — example session

PS C:\subs> gfortran --version
GNU Fortran (GCC) 13.2.0

PS C:\subs> gfortran -c umat_plasticity.f90 -o umat.o
(no output = success)

PS C:\subs> nm umat.o | grep umat
0000000000000000 T umat_

Open via the Terminal tab in the bottom panel, or Ctrl+`.

Testing

Single-Point RVE Test Runner

The heart of Jacobi. Each test generates a standalone Fortran program that calls your subroutine with known inputs, then checks the outputs against analytical solutions from the computational mechanics literature.

ℹ

How it works: Jacobi writes a Fortran driver program, compiles your subroutine together with it using gfortran, runs the binary, and parses tagged output lines. No Abaqus licence required. Total time: <3 seconds.

Configuring material parameters

Before running, set your material constants in the parameter bar at the top of the Tests tab. These values are injected into the Fortran driver's PROPS array.

Parameter	Description	Used by
`E`	Young's modulus (MPa)	UMAT, VUMAT, UEL
`nu`	Poisson's ratio	UMAT, VUMAT, UEL
`sy`	Initial yield stress (MPa). Also used as σ_ref in CREEP tests	UMAT, VUMAT, UHARD, CREEP
`Steps`	Number of strain increments in the sweep	All
`dEps/step`	Strain increment per step. Also sets DTIME = dEps × 1000 for CREEP	All

Testing → UMAT

UMAT Tests — 15 Physics Checks

Auto-detected when your file contains a SUBROUTINE UMAT signature. Tests cover elastic constants, plasticity, tangent stiffness, and thermodynamic consistency.

Elastic constants

ID	Test	What is tested	Reference
A	Uniaxial stiffness	`DDSDDE(1,1) · DSTRAN(1)` vs analytical C₁₁₁₁	de Souza Neto §4.3.2
B	Lateral confinement	`DDSDDE(1,2)` vs C₁₁₂₂ = Eν/((1+ν)(1−2ν))	de Souza Neto §4.3.2
C	Shear modulus	`DDSDDE(4,4)` vs G = E/2(1+ν)	de Souza Neto §4.3.2
D	Bulk modulus	Volumetric sum vs K = E/3(1−2ν)	de Souza Neto §4.3.1

Plasticity

ID	Test	What is tested	Reference
E	J2 hydrostatic neutrality	Hydrostatic compression stays elastic — `STRESS` components equal	de Souza Neto §8.2.1
F	Von Mises yield onset	√(3J₂) reaches `SYIELD` at the correct strain step	de Souza Neto §7.2.1
G	Elastic unloading slope	After yielding, unloading slope returns to C₁₁₁₁	de Souza Neto §7.3.1
K	Plastic incompressibility	tr(dε_p) = dεp₁₁+dεp₂₂+dεp₃₃ ≈ 0 in plastic regime	de Souza Neto §8.2.3
L	Yield surface return accuracy	f(σ,q) ≈ 0 after return-mapping — closest-point projection	Simo & Hughes §3.6.1
M	Flow direction normality	dεp₂₂/dεp₁₁ → −0.5 (Prandtl-Reuss, J2 associativity)	de Souza Neto §8.2.3
N	Drucker stability	σ:dεp ≥ 0 — non-negative plastic dissipation	de Souza Neto §6.2
O	Cyclic / Bauschinger	Compressive re-yield ≥ 0.9·σ_y0 — hardening sign on reverse load	de Souza Neto §12.2

Tangent stiffness

ID	Test	What is tested	Reference
H	Zero increment consistency	DSTRAN=0 → \|DSTRESS\| < 1e-8 and DDSDDE unchanged	de Souza Neto §7.5
I	DDSDDE FD check	DDSDDE vs dSTRESS/dDSTRAN via finite difference — consistent tangent	de Souza Neto §7.6.1, Simo & Hughes §3.5.3
J	DDSDDE symmetry	DDSDDE(i,j) = DDSDDE(j,i) — major symmetry of C	de Souza Neto §4.3.2

⚠

Test I (DDSDDE) is the most important. Returning the elastic stiffness in the plastic regime (the most common mistake) causes only linear convergence in Abaqus/Standard instead of quadratic. The test catches this immediately.

Testing → VUMAT

VUMAT Tests

For Abaqus/Explicit user material subroutines. VUMAT operates on a block of material points (NBLOCK), uses engineering shear strain (γ/2, not γ), and updates stressNew and enerInternNew in-place.

Test	Testing	Reference
A — Uniaxial elastic modulus	`stressNew(1)` = `stressOld(1)` + C₁₁₁₁·`strainInc(1)`	de Souza Neto §8.2.1
B — Shear modulus G	`stressNew(4..6)` from `strainInc(4..6)` — note: engineering shear γ/2	de Souza Neto §4.3.2
C — Bulk modulus K	Mean stress from volumetric strain	de Souza Neto §4.3.1
D — Elastic unloading slope	Slope of `stressNew(1)` during unloading after plastic loading = E	Lubliner §4.3
Energy — 2nd law	`enerInternNew(k)` ≥ `enerInternOld(k)` every increment	de Souza Neto §6.1

Testing → UEL

UEL Tests

Tests a 4-node plane-strain element (NDOFEL=8) with 2×2 Gauss integration. The three classical element validation tests from the FEM literature.

Test	Testing	Reference
A — AMATRX symmetry	max\|AMATRX(i,j) − AMATRX(j,i)\| < 1e-10	Cook, Malkus & Plesha §3.5
B — Rigid body null space	max\|RHS\| < 1e-6 under unit rigid-body translation (K·u_rigid = 0)	Bathe (1996) §4.3
C — Patch test	Element reproduces constant strain field exactly — completeness check	Irons & Razzaque (1972)
D — Positive diagonals	min(AMATRX(i,i)) > 0 — necessary for positive definiteness	Cook §2.4

Testing → UHARD

UHARD Tests

Tests the isotropic hardening law. UHARD must return SYIELD (current yield stress) and HARD(1) (dσ_y/dε̄_p, the tangent hardening modulus used in the return-mapping denominator).

Test	Testing	Reference
A — Initial yield	`SYIELD` at EQPLAS=0 equals σ_y0 from material parameters	Abaqus Manual §26.2.1
B — Monotonicity	SYIELD non-decreasing across all steps (Drucker stability)	Lubliner §5.3
C — HARD(1) FD consistency	`HARD(1)` vs central-difference (SYIELD(ε+δ)−SYIELD(ε−δ))/2δ	Simo & Hughes §3.5.2

Testing → CREEP

CREEP Tests

Tests against a Norton-Bailey power law (Δε_cr = A·σ̄ⁿ·DTIME). The driver sets A=1e-10, n=3, and uses σ_ref from the material parameter bar.

Test	Testing	Reference
A — Non-negativity	`DECRA(1)` ≥ 0 for any QTILD ≥ 0	Lemaitre & Chaboche §6.1.1
B — Norton-Bailey accuracy	`DECRA(1)` vs A·QTILD^n·DTIME — relative error < 0.01%	Abaqus Benchmark §3.2.6
C — Rate sensitivity	`DECRA(1)` at 2σ / DECRA(1) at σ = 2^n exactly	Bailey (1935)
D — Zero stress	`DECRA(1)` = 0 when QTILD = 0	Norton (1929)

      Fortran 90
      creep_power_law.f — example passing all 4 tests
    

SUBROUTINE CREEP(DECRA, DESWA, STATEV, SERD, EC, ESW, P, QTILD,
     &             TEMP, DTEMP, PREDEF, DPRED, TIME, DTIME, ...)
  IMPLICIT REAL*8(A-H,O-Z)

  A_CR = PROPS(1)   !  Coefficient A
  N_CR = PROPS(2)   !  Stress exponent n

  IF (QTILD .LE. 0.0D0) THEN
    DECRA(1) = 0.0D0   !  Test D: zero stress → zero creep
    DECRA(5) = 0.0D0
    RETURN
  END IF

  !  Norton-Bailey: Δε_cr = A * σ_bar^n * DTIME
  DECRA(1) = A_CR * QTILD**N_CR * DTIME
  DECRA(5) = A_CR * N_CR * QTILD**(N_CR - 1.0D0) * DTIME

END SUBROUTINE

Testing

Test History

Every test run is automatically saved with its material parameters, pass/fail results, and AI analysis text. Browse previous runs via the History tab.

Use Export Test Report in the AI Analysis section to generate a self-contained HTML report you can share with a colleague or attach to a paper submission. The report includes all charts, test results, and the AI diagnostic text.

AI

AI Diagnostics
Claude Sonnet

After running tests, click Analyze with AI. Jacobi sends only the failing tests to Claude Sonnet — with measured values, expected values, and canonical textbook references embedded — and receives a structured per-failure diagnosis.

What the AI receives

The prompt is built dynamically from your actual results. If 3 of 15 UMAT tests fail, the AI sees only those 3, each with:

The test name and its canonical reference (e.g. de Souza Neto §7.6.1 eq 7.7.5)
Your measured value vs the expected value
The suspect Fortran variable name (e.g. DDSDDE(1,1))

What the AI returns

For each failing test, the response is structured in three parts:

Example AI response — Test I fails (DDSDDE tangent)

## Test I — DDSDDE finite-difference check

Physical meaning:
  Your DDSDDE is the elastic stiffness C_e in all cases. In the plastic regime,
  Abaqus/Standard uses DDSDDE to form the global stiffness matrix. Returning C_e
  instead of the algorithmic consistent tangent C_ep causes only linear convergence
  instead of quadratic — your jobs will take 5-10× more iterations to converge.
  Reference: de Souza Neto §7.6.1, Simo & Hughes (1998) §3.5.3.

Most likely variable:
  DDSDDE(I,J) in your plastic branch. You are probably assigning the elastic
  values and not applying the plastic correction:
    C_ep = C_e - (C_e:n ⊗ n:C_e) / (H + n:C_e:n)

Suggested fix:
!  After computing DLAM and the flow direction RNORM, add:
  FACN = 1.0D0 / (HMOD + 3.0D0*GMOD)
  DO I = 1, NTENS
    DO J = 1, NTENS
      DDSDDE(I,J) = DDSDDE(I,J) - 2.0D0*GMOD*2.0D0*GMOD * FACN
     &                        * DEVS(I) * DEVS(J) / SMISES**2
    END DO
  END DO

✦

API key required. Add your Anthropic API key via the key icon in the toolbar. The key is stored locally and never sent to any server other than api.anthropic.com.

AI

AI Code Completion
Copilot-style

Ghost-text completions appear as you type in the editor, trained on Fortran patterns and Abaqus UMAT conventions. Press Tab to accept.

Example — typing starts a return mapping, AI completes

!  You type:
  SMISES = SQRT(1.5D0*

!  AI suggests (ghost text in grey):
  SMISES = SQRT(1.5D0*DOT_PRODUCT(DEVS,DEVS))

Completions are context-aware — they read the current subroutine type, declared variables, and surrounding code before generating.

Tools

Calibration Panel

Fit material parameters (E, ν, σ_y, H) to experimental data using least-squares optimisation. Paste stress-strain CSV data and the panel solves for the best-fit PROPS values.

1

Open the Calibration tab

Click Calibration in the bottom panel tabs.

2

Paste experimental data

Two-column CSV: strain (ε₁₁) in column 1, stress (σ₁₁, MPa) in column 2. The panel plots the data immediately.

3

Run calibration

Jacobi minimises ‖σ_UMAT(ε) − σ_exp(ε)‖ by varying the selected parameters. Results are shown alongside the experimental curve.

4

Copy to Tests

Click Apply to Tests to push the calibrated values into the Test Runner parameter bar, then rerun tests to verify the fitted model.

Tools

Step Inspector

Full per-increment state viewer. Shows every STRESS, STATEV, DDSDDE, DSTRAN, and STRAN value at every increment of the last test run. Useful for tracking down subtle integration errors that summary pass/fail metrics miss.

Step Inspector — increment 12 (plastic regime)

Step 12   EQPLAS = 0.0120   Phase: plastic

STRESS   [ 312.4  -156.2  -156.2   0.0   0.0   0.0 ]  MPa
DSTRAN   [ 1.00e-4  -4.30e-5  -4.30e-5  0  0  0 ]
STATEV   [ 0.0120  312.4  0.0  … ]
         (EQPLAS)  (SYIELD)

DDSDDE (6×6):
  221.4   110.7   110.7    0    0    0
  110.7   221.4   110.7    0    0    0
  ...
  ← plastic tangent, not elastic (C_ep ≠ C_e) ✓

Tools

STATEV Manager

Documents your solution-dependent state variables. Map each STATEV(i) slot to a name, unit, and description. The mapping is saved with the project and shown in the Step Inspector.

STATEV mapping — J2 plasticity example

STATEV(1)  EQPLAS   Equivalent plastic strain          [—]
STATEV(2)  SYIELD   Current yield stress               [MPa]
STATEV(3)  ALPHA1   Backstress component 11            [MPa]
STATEV(4)  ALPHA2   Backstress component 22            [MPa]
STATEV(5)  ALPHA3   Backstress component 33            [MPa]
STATEV(6)  ALPHA4   Backstress component 12            [MPa]

Tools

Hardening Curve Editor

Visual editor for piecewise-linear hardening data. Plot your σ_y vs ε̄_p table, interpolate, and export directly as Fortran DATA statements or as Abaqus *PLASTIC keyword input.

Tools

.inp File Viewer

Parse and display Abaqus .inp files — nodes, elements, sections, boundary conditions, steps, and material blocks. Useful for checking that the UMAT is being called with the correct PROPS, NSTATV, and NTENS values.

Tools

Convergence Log

Parses Abaqus .msg and .dat files to visualise Newton iteration counts, residual norms, and cutbacks. Helps correlate test failures (especially DDSDDE) with real-simulation convergence behaviour.

Reference

Subroutine Reference

All Abaqus subroutine types recognised by Jacobi. Hover any subroutine name in the editor to see its full argument list and description.

Abaqus/Standard

UMAT

UMATHT

UEL

UELMAT

UHARD

CREEP

DLOAD

DFLUX

UFIELD

USDFLD

URDFIL

UEXTERNALDB

UTRS

UEXPAN

Abaqus/Explicit

VUMAT

VUMATHT

VUEL

VDLOAD

VDFLUX

VDISP

VEXTERNALDB

Reference

Keyboard Shortcuts

File & Editor

Open File

CtrlO

Save File

CtrlS

Command Palette

CtrlShiftP

Zoom In / Out / Reset

Ctrl+ Ctrl- Ctrl0

Next diagnostic squiggle

F8

Build & Test

Build (compile + run)

F5

Open Terminal

Ctrl`

Editor (Monaco standard)

Find / Replace

CtrlF CtrlH

Go to Line

CtrlG

Toggle comment

Ctrl/

Select all occurrences

CtrlShiftL

Format (indent)

ShiftAltF

Reference

Export & Reports

After running tests and clicking Analyze with AI, use Export Test Report to generate a self-contained HTML document containing:

Material parameters used
All 35 test results with pass/fail and measured values
Charts for every test with a numeric output
The full AI diagnostic text, including suggested Fortran fixes
Literature references for each test

The HTML file is fully self-contained (no external dependencies) and can be attached to a paper, sent to a reviewer, or archived alongside the subroutine source.

Feature Reference

What is Jacobi?

Get running in 60 seconds

Fortran Editor

Syntax highlighting

Static analysis

Themes

Command Palette

Subroutine Templates

Build System

Running a build

Error navigation

Integrated Terminal

Single-Point RVE Test Runner

Configuring material parameters

UMAT Tests — 15 Physics Checks

Elastic constants

Plasticity

Tangent stiffness

VUMAT Tests

UEL Tests

UHARD Tests

CREEP Tests

Test History

AI Diagnostics Claude Sonnet

What the AI receives

What the AI returns

AI Code Completion Copilot-style

Calibration Panel

Step Inspector

STATEV Manager

Hardening Curve Editor

.inp File Viewer

Convergence Log

Subroutine Reference

Abaqus/Standard

Abaqus/Explicit

Keyboard Shortcuts

File & Editor

Build & Test

Editor (Monaco standard)

Export & Reports

AI Diagnostics
Claude Sonnet

AI Code Completion
Copilot-style