Linear vs Non-Linear Structural Analysis

NON-LINEAR ANALYSIS IS NOT AUTOMATICALLY MORE ACCURATE

There is a common assumption that non-linear analysis is inherently superior to linear analysis — that if the solver has a non-linear option, turning it on produces a "better" result. This is a dangerous misconception. Non-linear analysis opens the door to more realistic physics, but it also opens the door to more ways to get the wrong answer. A linear analysis has a single solution; a non-linear analysis has a solution path that may converge, may converge to the wrong branch, or may fail to converge at all. The result of a non-linear analysis looks every bit as convincing as a linear result — a colourful contour plot, a deflection value, a reaction summary — but it can be wrong in ways that a linear analysis cannot be. Non-linear analysis requires better inputs, better solver settings, and better validation. Used appropriately, it is the right tool; used carelessly, it is a more sophisticated way to be wrong.

Non-linear analysis is a tool for problems where the physics is genuinely non-linear. It is not an upgrade that automatically improves accuracy. If the problem is linear, a linear analysis is the correct choice — faster, simpler, and no less accurate.

The Three Sources of Non-Linearity

Structural non-linearity has three distinct sources. A given problem may involve one, two, or all three. Understanding which sources are active is the first step in deciding whether non-linear analysis is needed.

• MATERIAL non-linearity: the stress-strain relationship is not linear — plasticity, hyperelasticity, viscoelasticity, damage, creep
• GEOMETRIC non-linearity: the deformation is large enough that the geometry changes significantly — large displacement, large rotation, stress stiffening, snap-through, buckling
• CONTACT/BOUNDARY non-linearity: the support or load transfer changes with deformation — contact opening/closing, sliding, separation, changing constraints

linear-nonlinear

Material Non-Linearity

Material non-linearity occurs when the stress-strain relationship departs from linear elasticity. The most common form is plasticity — the material yields and deforms permanently when the stress exceeds the yield point. Plasticity is essential for predicting ultimate strength, crash behaviour, and permanent deformation. Other forms include hyperelasticity (rubber, elastomers — large elastic strain with a non-linear stress-strain curve), viscoelasticity (polymers — time-dependent response), creep (metals at elevated temperature — time-dependent permanent deformation), and damage (composites, concrete — progressive degradation of stiffness). Material non-linearity requires accurate material data — a yield strength, a hardening curve, a creep law — and the result is sensitive to the quality of that data. A non-linear material model with guessed or extrapolated data can be less accurate than a linear model with a conservative allowable.

• Plasticity: yielding and permanent deformation — requires yield strength and hardening curve
• Hyperelasticity: large elastic strain (rubber, elastomers) — requires hyperelastic material constants
• Viscoelasticity: time-dependent response (polymers) — requires relaxation or creep data
• Creep: time-dependent permanent deformation (elevated temperature metals) — requires creep law and constants
• Damage: progressive stiffness degradation (composites, concrete) — requires damage evolution parameters

Geometric Non-Linearity

Geometric non-linearity occurs when the deformation is large enough that the changing geometry affects the response. In a linear analysis, the equilibrium equations are written on the undeformed geometry. In a geometrically non-linear analysis, the equations are written on the deformed geometry, updated as the load is applied. This captures two important effects. First, stress stiffening: as a membrane or thin shell deforms, the out-of-plane displacement generates in-plane tension that stiffens the structure — a catenary effect. Second, large rotation: a component that rotates significantly changes its load direction and load path. Both effects can be stabilising or destabilising. A pressure-loaded panel that develops membrane tension is stiffer than the linear prediction (stabilising). A column that approaches its buckling load becomes increasingly sensitive to imperfection (destabilising). A shallow arch that snaps through to a new configuration (snap-through buckling) has no linear analogue at all.

• Large displacement: deflection is a significant fraction of the structural dimension — geometry changes affect the response
• Large rotation: the component rotates enough that the load direction relative to the component changes
• Stress stiffening: out-of-plane deformation generates in-plane tension — stabilising for membranes and thin shells
• Snap-through: the structure jumps to a new equilibrium configuration — no linear analogue
• Buckling: eigenvalue buckling is a linearised prediction; non-linear buckling traces the actual post-buckling path

Contact and Boundary Non-Linearity

Contact is the third source of non-linearity. When two surfaces come into contact, the load transfer between them depends on whether they are open, closed, sticking, or sliding — and this status changes as the load is applied and the structure deforms. Contact is non-linear because the boundary conditions themselves change during the analysis. A joint that is open at the start of the analysis may close under load; a bolted interface that is clamped at the start may open under bending; a sliding contact may transition to sticking or to separation. Each change of contact status changes the stiffness matrix, requiring a new solution. This is why contact analysis is computationally expensive and convergence-sensitive — the solver must detect and adapt to changing boundary conditions at every step.

• Contact opening/closing: surfaces separate or come together as the structure deforms
• Sliding vs sticking: friction determines whether contact surfaces slide or stick — and this can change during the analysis
• Changing support: a support that is active under one load direction may lift off under another
• Preload changes: bolt preload is applied in one step; external loads in the next; the contact status at the clamped interface evolves

Linear Superposition and Why It Fails

In a linear analysis, the principle of superposition holds: the response to a combination of loads is the sum of the responses to each load applied separately. This is why linear analysis is so efficient — a single stiffness matrix inversion gives the response to any load vector, and load cases can be combined after the solve. In a non-linear analysis, superposition does not hold. The stiffness matrix changes with the load level, so the response to load A plus load B is not the response to A plus the response to B. Each load combination must be solved as a separate analysis. This has practical consequences: a non-linear analysis cannot use superposition to combine load cases, and a non-linear dynamic analysis cannot use modal superposition.

    Linear:        K · u_A  =  F_A      →    u_A  =  K⁻¹ · F_A
                  K · u_B  =  F_B      →    u_B  =  K⁻¹ · F_B
                  K · u_{A+B}  =  F_A + F_B  →  u_{A+B}  =  u_A + u_B   ✓  (superposition holds)

    Non-linear:    K(u) · u  =  F       →    u  =  [K(u)]⁻¹ · F
                  K depends on u, so:
                  u_{A+B}  ≠  u_A + u_B   ✗  (superposition fails)

    Each load combination must be solved as a separate non-linear analysis.

If your analysis is non-linear, you cannot combine load cases by superposition. Each critical load combination must be solved separately. This can multiply the computational cost by the number of load combinations.

Incremental Solution and Newton-Raphson

A non-linear analysis cannot be solved in a single matrix inversion. The solver applies the load in increments, and at each increment it iterates to find the equilibrium configuration. The most common iterative method is Newton-Raphson: at each iteration, the solver forms the tangent stiffness matrix at the current state, solves for the displacement correction, updates the displacement, and checks equilibrium. If equilibrium is satisfied within tolerance, the increment converges and the solver moves to the next. If not, it iterates again. Newton-Raphson converges quadratically when it works — the residual drops rapidly with each iteration. But it can fail to converge if the load step is too large, if the material model is discontinuous, if contact status oscillates, or if the structure is near a instability point (buckling, snap-through).

• Load increments: the total load is applied in steps, not all at once
• Newton-Raphson: at each iteration, form tangent stiffness, solve for correction, update, check equilibrium
• Convergence: quadratic when it works — residual drops rapidly
• Failure: oscillating contact, discontinuous material, too-large step, near-instability

Convergence Challenges

Non-linear convergence is not guaranteed. The most common causes of convergence failure are: contact chatter (surfaces oscillate between open and closed, preventing equilibrium), material instability (a softening material model can produce a non-positive-definite stiffness matrix), large load steps (the increment is too large for Newton-Raphson to converge), and buckling or snap-through (the structure passes through a instability point where the tangent stiffness becomes singular). The remedies include: smaller load steps, automatic stabilisation (artificial damping to bridge instabilities), contact stiffness adjustments (reducing penalty stiffness to reduce chatter), and arc-length methods (which trace the equilibrium path through a limit point by controlling both load and displacement). Diagnosing a convergence problem requires reading the solver log — the residual force, the displacement correction, and the contact status at each failed iteration.

A non-linear analysis that fails to converge is not a "conservative" result. It is no result at all. Diagnose the convergence failure by reading the solver log: where did the residual stop decreasing? Which contact pair is oscillating? Which element is distorting? Fix the cause before drawing any conclusion.

Examples: When Non-Linearity Matters

Three examples illustrate when non-linear analysis is essential and when it is not.

• Bolt contact: a bolted joint with contact and friction is non-linear — the load transfer through the joint depends on contact status and slip. A linear analysis with bonded contact gives a single, potentially wrong load path. Non-linear contact is essential.
• Large-deflection panel: a thin panel under pressure develops membrane stress as it deflects. A linear analysis underestimates the stiffness and overestimates the deflection. Geometric non-linearity is essential.
• Yielding bracket: a bracket loaded beyond yield redistributes stress through plastic deformation. A linear analysis overestimates the peak stress and cannot predict the ultimate load. Material non-linearity is essential for ultimate strength.
• When linear is enough: a thick steel bracket under a moderate load, well below yield, with small deflection and no contact — a linear analysis is accurate, fast, and the correct choice.

When to Use Non-Linear Analysis

The decision to use non-linear analysis should be driven by the physics of the problem, not by the availability of a non-linear solver. If the stress is well below yield, the deflection is small, and the contact and support conditions do not change, a linear analysis is the correct choice. If any of these conditions is violated, non-linear analysis is needed. The cost of non-linear analysis — in solution time, in input preparation, in validation — should be justified by the physics.

• Is the stress above yield anywhere in the model? — If yes, material non-linearity is needed for ultimate strength
• Is the deflection a significant fraction of a structural dimension? — If yes, geometric non-linearity is needed
• Does the load transfer change with deformation (contact, support lift-off)? — If yes, contact/boundary non-linearity is needed
• Is the structure near a buckling or snap-through instability? — If yes, geometric non-linearity with arc-length methods is needed
• Are the material properties time- or rate-dependent? — If yes, viscoelasticity or creep non-linearity is needed
• If none of the above — is a linear analysis sufficient? — If so, it is the correct choice: faster, simpler, and no less accurate

Key takeaways

• Non-linearity has three sources: material (plasticity, hyperelasticity, damage), geometric (large displacement, large rotation, stress stiffening), and contact/boundary (opening, closing, sliding, changing support).
• Linear superposition does not hold in non-linear analysis — doubling the load does not double the response, and combining load cases requires re-solving, not adding.
• Non-linear analysis solves incrementally, using Newton-Raphson or similar methods to iterate to equilibrium at each load step — convergence is not guaranteed.
• NON-LINEAR ANALYSIS IS NOT AUTOMATICALLY MORE ACCURATE — it requires better material data, better contact definitions, better boundary conditions, and better validation than a linear analysis.
• A non-linear analysis that fails to converge is not a "conservative" result — it is no result at all. Diagnose and fix the convergence problem before drawing any conclusion.
• Geometric non-linearity (large displacement) can be stabilising (stress stiffening) or destabilising (buckling, snap-through) — the effect depends on the structure and the load.
• The decision to use non-linear analysis should be driven by the physics of the problem, not by the availability of a non-linear solver.