Comprehensive Guide on Nonlinear FEA Using the

Newton–Raphson Method
Dr. Iman Salehinia
April 14, 2025
MechCADemy Youtube Channel: https://www.youtube.com/@MechCADemy-e9y
NIU Webpage: https://www.niu.edu/ceet/about/directory/salehinia-iman.shtml

Contents
1 Overview of the Newton–Raphson Method 2
1.1 Single-Variable Formulation and Example . . . . . . . . . . . . . . . . . . . 2
1.2 Multi-Variable Newton–Raphson . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Newton–Raphson in Finite Element Analysis 4

2.1 From Linear FEA to Nonlinear FEA . . . . . . . . . . . . . . . . . . . . . . 4
2.2 Tangent Stiffness Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Nonlinear Spring Example (Detailed Iterations) . . . . . . . . . . . . . . . . 4
2.4 Convergence Criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3 Step-by-Step Procedure for Nonlinear FEA 5

3.1 Global Roadmap: Total Lagrangian + Newton–Raphson . . . . . . . . . . . 5
3.2 Detailed Iteration Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

4 Kinematic Measures in a Total Lagrangian Approach 6

4.1 Deformation Gradient F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.2 Green–Lagrange Strain E . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.3 Strain–Displacement Matrix B . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.4 Illustrative 2D Bilinear Example . . . . . . . . . . . . . . . . . . . . . . . . . 7

5 Second Piola–Kirchhoff (2PK) Stress and Internal Force 9

5.1 Why the 2PK Stress? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
5.2 Constitutive Modeling (Elastic,Plastic) . . . . . . . . . . . . . . . . . . . . . 9
5.3 Computation of Fint . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
5.4 Relation to the Cauchy Stress . . . . . . . . . . . . . . . . . . . . . . . . . . 9

6 Putting It All Together 10

6.1 Full Iterative Algorithm Summary . . . . . . . . . . . . . . . . . . . . . . . . 10
6.2 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

7 Final Reflections: Why the Roadmap Matters 11

8 Note on Formulas vs. Derivations 11

1
1 Overview of the Newton–Raphson Method
The Newton–Raphson (NR) method is a powerful root-finding (or system-solving) algo-
rithm that leverages local linearization (first-order Taylor expansions) to obtain fast conver-
gence,provided the initial guess is close enough and the Jacobian is well-conditioned.

1.1 Single-Variable Formulation and Example

When you want to solve a single equation

f (x) = 0,

the Newton–Raphson iteration reads:

f x(k)
x(k+1) = x(k) − ′ (k)
.
f x

Example: Solve √
f (x) = x2 − 2 = 0 −→ x = ± 2.
The derivative is f ′ (x) = 2x.Plugging into the formula:

x(k)2 − 2
x(k+1) = x(k) − .
2 x(k)

Table 1: Iteration table for the single-variable example x2 − 2 = 0,starting from x(0) = 2.0.
f ′ x(k)



Iter. x(k) f x(k) x(k+1)
0 2.0000 2.0000 4.0000 1.5000
1 1.5000 0.2500 3.0000 1.4167
2 1.4167 0.00694 2.8333 1.4142
3 1.4142 0.00006 2.8284 1.4142
√
This rapidly converges to 2 ≈ 1.4142.

1.2 Multi-Variable Newton–Raphson

When the unknown vector is x ∈ Rn and you have a system

f (x) = 0,

the Jacobian J(x) is the matrix of partial derivatives:

 ∂f1 ∂f1

∂x1
· · · ∂xn
J(x) =  ... . . . ...  .
 
∂fn ∂fn
∂x1
··· ∂xn

2
For a 2D system (x, y),this simplifies to:
∂f1 ∂f1
!
∂x ∂y
J(x, y) = ∂f2 ∂f2
.
∂x ∂y

The Newton–Raphson iteration in vector form is:

−1
x(k+1) = x(k) − J x(k) f x(k) .


Example System: (
x + y − 4 = 0,
x2 + y 2 − 20 = 0.
Define    
x+y−4 1 1
f (x, y) = , J(x, y) = .
x2 + y 2 − 20 2x 2y

Sample Calculation (Iteration 0):

(x(0) , y (0) ) = (3, 1).

Then      
3+1−4 0 1 1
f (3, 1) = = , J(3, 1) = .
9 + 1 − 20 −10 6 2
The determinant is (1)(2) − (1)(6) = −4, so
 
−1 −1/2 1/4
J (3, 1) = .
3/2 −1/4

Hence,  (1)        
x 3 −1/2 1/4 0 5.5
= − = .
y (1) 1 3/2 −1/4 −10 −1.5

Table 2: Newton–Raphson iteration steps for x + y − 4 = 0, x2 + y 2 − 20 = 0.

(k) (k)
k (x(k) , y (k) ) (f1 , f2 ) (x(k+1) , y (k+1) )
0 (3.0, 1.0) (0, −10) (5.5, −1.5)
1 (5.5, −1.5) (0, 12.5) (4.6071, −0.6071)

Iteration Table √(Final √ Results): Continuing further iterations converges to one real
solution,e.g.
√ (2
√ + 6, 2 − 6).Different initial guesses might converge to the other solution
(2 − 6, 2 + 6).

3
2 Newton–Raphson in Finite Element Analysis
2.1 From Linear FEA to Nonlinear FEA
In linear finite element analysis,we often have

Fext = K a,

where

• a is the nodal displacement vector,

• K is the constant global stiffness matrix,

• Fext is the external load vector.

One inverts K once to get a = K−1 Fext .

However,in nonlinear FEA,the internal force Fint (a) depends on a nonlinearly:

Fint (a) ̸= K a (no longer constant K),

thus requiring iterative solution of

Fext − Fint (a) = 0.

2.2 Tangent Stiffness Matrix

Within each iteration,we linearize Fint (a) around the current guess a(i) :

∂Fint
Fint a(i+1) ≈ Fint a(i) + a(i+1) − a(i) .




∂a a(i)

Define
(i) ∂Fint
KT = (the tangent stiffness).
∂a a(i)
The global Newton–Raphson iteration in FEA is:
(i)
KT ∆a(i) = Fext − Fint a(i) , a(i+1) = a(i) + ∆a(i) .

2.3 Nonlinear Spring Example (Detailed Iterations)

Consider a 1D nonlinear spring where

Fint (u) = k u + α u3 ,

with constants k and α.Suppose an external force Fext is applied.We want Fint (u) = Fext .The
tangent stiffness is
d
k u + α u3 = k + 3 α u2 .


KT (u) =
du
4
 Table 3: Iteration table for a 1D nonlinear spring.
Iter. u(i) Fint (u(i) ) R(i) = Fext − Fint KT (u(i) ) ∆u(i) u(i+1)
0 1.000 11.0 9.0 13.0 9/13=0.692 1.692
1 1.692 21.75 -1.75 18.59 -1.75/18.59=-0.094 1.598
2 1.598 20.06 -0.06 17.66 -0.06/17.66=-0.0034 1.5946
3 1.5946 19.999 0.001 17.63 ≈ 6 × 10−5 1.59466

Example: Let k = 10 N/m,α = 1 N/m3 ,and Fext = 20 N.Solve k u + α u3 = 20.

Converged solution: u ≈ 1.5946 m.

2.4 Convergence Criteria

In practice,one sets tolerances on:
• Residual Norm: ∥R(i) ∥ = ∥Fext − Fint ∥.
• Displacement Increment Norm: ∥∆a(i) ∥.
∥∆a ∥ (i)
• Relative Changes: e.g., ∥a(i+1) ∥ < ϵ.

Newton–Raphson typically converges quadratically when close to the solution and the tangent
matrix is well-conditioned.

3 Step-by-Step Procedure for Nonlinear FEA

3.1 Global Roadmap: Total Lagrangian + Newton–Raphson
When dealing with large deformations or material nonlinearities,a Total Lagrangian
formulation (where everything is measured in the reference configuration) is common.Typical
high-level steps:
1. Define geometry,material model (e.g. elastic,elastoplastic,hyperelastic),mesh,boundary
conditions,and loads.
2. Possibly subdivide the total load into increments (load step 1,2,etc.).
3. Initialize all unknown displacements at zero or a known approximation.
4. Newton–Raphson Loop for each load step:
• Compute internal forces and tangent stiffness.
• Solve for displacement increments.
• Update displacements.
• Check convergence.
5. Post-process results: final displacements,stress distribution,etc.

5
3.2 Detailed Iteration Flow
Adapted from a Step-by-Step Guide to Solving Nonlinear FEA Problems,each load step n
typically does:
(n)
1. Apply Fext (this is not the full external load,but rather the incremental or reduced
portion for this load step).

2. Set iteration index i = 0.Use last converged a from the previous load step as the initial
guess.

3. Compute the global residual

(n)
R(i) = Fext − Fint a(i) .

(i)
4. Assemble global tangent stiffness KT .

• Requires element-level computations: the deformation gradient F,the stress mea-

sure S,etc.

5. Check if ∥R(i) ∥ < tolerance.If not converged:

(i)
KT ∆a(i) = R(i) , a(i+1) = a(i) + ∆a(i) .

6. Repeat until convergence.

7. Proceed to next load step (or finish if final load is reached).

4 Kinematic Measures in a Total Lagrangian Approach

A central piece of large-deformation FEA is computing:

• The deformation gradient F,

• The Green–Lagrange strain E,

• The strain–displacement matrix B.

All are evaluated with respect to the reference (initial) configuration.

Note: The following example for F and E simply illustrates a sample computation at
one Gauss point.In actual FEA codes,you would repeat this at each Gauss point of each
element.We also assume the reader has basic knowledge of isoparametric mapping,shape
functions,and simple stress analysis in FEA.

6
4.1 Deformation Gradient F
For a 2D (or 3D) element,let:
• (Xi , Yi ) be the coordinates of node i in the reference configuration,
• (xi , yi ) be the coordinates in the current (deformed) configuration:
xi = X i + u i , yi = Yi + vi .

Within an element,the position x(r, s) is obtained by shape functions Ni (r, s) (where

(r, s) ∈ [−1, 1] × [−1, 1]):
4
X 4
X
x(r, s) = Ni (r, s) xi , y(r, s) = Ni (r, s) yi .
i=1 i=1

Similarly for X(r, s) and Y (r, s) in the reference configuration.

4.2 Green–Lagrange Strain E

Once F is known,the Green–Lagrange strain is
1
FT F − I .


E= 2

Unlike small-strain ϵ = 12 (∇u + ∇uT ),E captures large rotations and stretches.

4.3 Strain–Displacement Matrix B

In a total Lagrangian element formulation,we express variations in E (or other strain mea-
sures) in terms of variations in nodal displacements δu.The matrix B collects all partial
derivatives so that:
δE = B δu.
For a 2D bilinear quad (4 nodes,each with 2 DOFs),B is 3 × 8.The exact form depends on
partial derivatives of the shape functions w.r.t. the reference coordinates (X, Y ).

4.4 Illustrative 2D Bilinear Example

Let’s consider a single 4-node element:
• Reference coordinates:
(X1 , Y1 ) = (0, 0), (X2 , Y2 ) = (2, 0), (X3 , Y3 ) = (2, 1), (X4 , Y4 ) = (0, 1).

• Current (deformed) coordinates: at iteration,

(xi , yi ) = (Xi + ui , Yi + vi ).
For example,
(u1 , v1 ) = (0.10, 0.00) → (x1 , y1 ) = (0.10, 0.00), (u2 , v2 ) = (0.15, 0.00) → (x2 , y2 ) = (2.15, 0.00),
etc.

7
 • Shape functions for a 4-node bilinear quad in (r, s):

N1 (r, s) = 14 (1−r)(1−s), N2 (r, s) = 41 (1+r)(1−s), N3 (r, s) = 14 (1+r)(1+s), N4 (r, s) = 41 (1−r)(1

∂(x,y) ∂(X,Y )
Sample Computation of ∂(r,s)
and ∂(r,s)
: - First,note that

4
X 4
X
x(r, s) = Ni (r, s) xi , y(r, s) = Ni (r, s) yi .
i=1 i=1

Therefore,
4 4
∂x X ∂Ni ∂x X ∂Ni
= xi , = xi ,
∂r i=1
∂r ∂s i=1
∂s
and similarly for y(r, s).
- For X(r, s) and Y(r, s) in the reference domain:
4
X 4
X
X(r, s) = Ni (r, s) Xi , Y (r, s) = Ni (r, s) Yi .
i=1 i=1

Then
4 4
∂X X ∂Ni ∂X X ∂Ni
= Xi , = Xi ,
∂r i=1
∂r ∂s i=1
∂s
(and similarly for Y ).
- Example partial derivatives: Suppose (r, s) = (0.57735, 0.57735),then
∂N1 ∂N1
= − 41 (1 − s), = − 14 (1 − r), ...
∂r ∂s
Evaluate these at r = s = 0.57735,multiply by the current node coords (xi , yi ) or reference
coords (Xi , Yi ),and sum.
- If the final numeric result is,for instance,
   
∂(x, y) 1.04572 −0.00528 ∂(X, Y ) 1.0 0.0
= , = ,
∂(r, s) 0.00394 0.52394 ∂(r, s) 0.0 0.5
∂(X,Y )
you then proceed to invert ∂(r,s)
to form F.

Final Steps to Get F and E:

h i−1 1.0 0.0 ∂(x, y) h ∂(X,Y ) i−1

1.04572 −0.01056

∂(X,Y )
∂(r,s)
= , F= = .
0.0 2.0 ∂(r, s) ∂(r,s) 0.00394 1.04788

Then
E = 12 (FT F − I).
Hence you capture the element’s local deformation measures in a Total Lagrangian frame-
work.

8
5 Second Piola–Kirchhoff (2PK) Stress and Internal
Force
5.1 Why the 2PK Stress?
In finite strain mechanics,the usual (Cauchy) stress σ is measured in the current (de-
formed) configuration.However,in a Total Lagrangian framework,all integrations are done
over the reference volume V0 .The 2nd Piola–Kirchhoff stress S is the natural stress mea-
sure conjugate to the Green–Lagrange strain E under the reference domain:
Z
δWint = SδE dV0 .
V0

5.2 Constitutive Modeling (Elastic,Plastic)

• Linear Elastic (finite-strain sense):

S = C : E,

where C is the 4th-order elasticity tensor.

• Bilinear Plastic example: once the equivalent strain ε̄ surpasses yield,you define S(E)
with a piecewise formula.

5.3 Computation of Fint

In the Total Lagrangian approach,the internal force at the element level is
Z
(e)
Fint = BT S dV0 ,
(e)
V0

where B is derived from reference shape function derivatives.Summing over all elements
yields the global Fint ,used in the Newton–Raphson equations.

5.4 Relation to the Cauchy Stress

If you want to interpret the true (physical) stress in the deformed configuration,convert S
to σ by
1
σ= F S FT .
det(F)
Hence,S is mostly for the internal (reference-domain) calculations; σ is for final post-
processing.

9
6 Putting It All Together
6.1 Full Iterative Algorithm Summary
A unified step-by-step for Nonlinear FEA in the Total Lagrangian setting:

1. Preprocessing:

• Define geometry (reference coords {Xi , Yi }),element connectivity,shape functions

Ni (r, s),
• Specify material properties (E, ν) or advanced models,
• Choose boundary conditions and load increments.

2. Initialize:

• Set a(0) = 0,
• Decide on convergence tolerances.

3. For Each Load Step n = 1, . . . , N :

(n)
• Apply or update Fext (the reduced portion for step n),
• Let a(0) = converged solution from the previous load step.

4. Newton–Raphson Iteration:

(a) Compute current nodal positions x(i) = X + a(i) ,

(b) For each element:
 ∂(X,Y ) −1
• Evaluate ∂(x,y)
∂(r,s)
and ∂(r,s)
to get F,
• Compute E = 21 (FT F − I),
• Obtain S from the constitutive law,
• Form element B and compute
Z Z
(e) (e)
Fint = B S dV0 , KT = BT Ctangent B dV0 .
T

(i) (i)
(c) Assemble global Fint and KT ,
(n) (i)
(d) Evaluate R(i) = Fext − Fint ,
(e) Check ∥R(i) ∥< tol.If not converged,
(i)
KT ∆a(i) = R(i) , a(i+1) = a(i) + ∆a(i) ,

increment i → i + 1 and repeat.

5. Convergence:
(n+1)
• Once converged for load step n,proceed to Fext .

10
 6. Postprocessing:

• Evaluate final displacements,2PK stress S,or convert to Cauchy σ,

• Plot stress contours,deformed shape,etc.

6.2 Concluding Remarks

By merging the power of the Newton–Raphson iteration with the Total Lagrangian formu-
lation, you can:

• Handle large deformations (finite strain),

• Incorporate material nonlinearities,

• Achieve robust convergence with suitable load increments and guesses.

7 Final Reflections: Why the Roadmap Matters

By following each step in this **Total Lagrangian + Newton–Raphson** roadmap, you gain
insight into *precisely what* your FEA software does when you press the “solve” button.
Instead of treating the solver as a black box, you now understand:
- **How the Newton–Raphson algorithm** iteratively updates displacements to reduce
the global residual, - **Why each element** has its own internal force contributions, and
how these assemble into a global system, - **What the “tangent stiffness”** (or Jacobian)
represents, and why it’s constantly updated in nonlinear problems, - **Why residuals** can
be mapped back to each element or node, allowing you to visualize equilibrium imbalances.
Armed with this knowledge, you can **build better models** and **interpret solver
results** more confidently. For instance, if you see that the residual stalls during iteration,
you can infer that your stiffness matrix might be poorly conditioned—or that you need more
careful load stepping. If you notice large element-level residuals in certain areas, you can
refine the mesh or verify boundary conditions there. Ultimately, **understanding the “why”
behind each solver step** helps you debug issues, explore advanced material models, and
trust your simulation results.
In short, when you know what happens behind the curtain, you’re empowered to **create
more accurate, stable, and well-validated nonlinear FEA models**, whether in Ansys or any
other solver. This deeper understanding will eventually help you become a more confident
CAE engineer!

8 Note on Formulas vs. Derivations

In this guide, we have **presented** the core formulas for the tangent stiffness matrix KT ,
the internal force vector Fint , and the deformation measures F and E **without** delving
into the detailed derivations that arise from:

• The principle of virtual work,

11
 • The principle of minimum potential energy,

• Calculus of variations,

• Continuum mechanics.

Readers who wish to see **how** these integral forms originate from the fundamental equa-
tions of solid mechanics and advanced finite-element theory are encouraged to consult more
specialized references or textbooks. Understanding the deeper theoretical foundations will
**further strengthen** your grasp of **why** these formulas take the forms they do.

End of Document.

12