What does first-order refer to?

The linear approximation of the transformed limit-state boundary at the design point.

Is the design point necessarily the input mean?

No. It is found on the boundary through a constrained optimization problem.

Is Phi(-beta) always the correct orientation?

No. The failure side and whether the origin lies in failure must be considered.

AI Glossary / Statistics and probability / Advanced

First-order reliability method

Definition

The first-order reliability method, or FORM, estimates a threshold-event probability by approximating a limit-state boundary with a tangent hyperplane at a specially chosen design point. In the common standard-normal formulation, the design point is the nearest point on the transformed boundary to the origin. Its distance defines a reliability index used in the probability calculation. FORM is an approximation method for a specified probabilistic model, not a safety certification or a guarantee that the computed failure probability is exact.

What it means

Reliability problems often involve several uncertain inputs and an expensive response model. The event of interest may occur only in a small region of input space. FORM replaces the difficult boundary near an important point with a simpler linear one whose probability can be calculated readily. The transformation to standard space matters because distance in arbitrary physical units would not reflect the input distributions. The method differs from simply linearizing a response at its mean input values, which is associated with mean-value moment approximations.

How it works

Define a limit-state function g(x) and which side represents failure. Specify the joint input distribution, including dependence, and use a suitable transformation to independent standard normal coordinates u. Find the point u-star minimizing its distance to the origin subject to the transformed limit state being zero. Linearize the transformed boundary there and calculate the normal probability of the resulting half-space. If the origin is on the safe side and beta is the design-point distance, the usual estimate is Phi(-beta). If the origin lies in failure, the orientation changes and the complementary large probability is required. Check optimization convergence, gradients and possible competing design points.

Three practical examples

Reliability analysis: approximate a threshold-exceedance probability for a model with uncertain loads and material properties.
Computational screening: identify regions of input space that warrant more detailed probability evaluation.
Sensitivity review: study how modeled uncertainty and local boundary direction influence a reliability estimate.

Think of it this way

Think of replacing a curved frontier with a flat line at its most relevant nearby point, after rescaling the map to reflect probability. Measuring the distance to that line makes the probability calculation easier. The shortcut becomes less trustworthy if the frontier bends sharply or has several important disconnected parts.

Worked example

Let U1 and U2 be independent standard normal variables and define the hypothetical safe margin g = 10 - 3U1 - 4U2. Failure is g less than or equal to zero, so the boundary is 3U1 + 4U2 = ten. The closest boundary point to the origin is (1.2, 1.6), verified because 3 x 1.2 + 4 x 1.6 = ten and the point lies in the normal direction (3,4). Its distance is sqrt(1.2 squared + 1.6 squared) = two. FORM therefore gives Phi(-2), approximately 0.0227501. Here the result is exact because the limit state is already linear in standard normal space. Independently, 3U1 + 4U2 is normal with mean zero and standard deviation five, so exceeding ten has the same tail probability. This benchmark does not establish accuracy for a curved real-world boundary.

Illustrative case study

Fictional case: Hemlock Engineering uses FORM to screen several simulated component configurations. Analysts verify the implementation against a linear benchmark, then inspect convergence from different starting points for the detailed model. Where the failure boundary appears strongly curved, they compare the estimate with additional numerical checks. The report distinguishes model assumptions from algorithm accuracy and leaves acceptance decisions to the applicable engineering review. This invented case illustrates disciplined use of an approximation and does not recommend a particular reliability threshold or claim a design has been validated.

Common misconceptions and mistakes

Do not use physical-space distance as a reliability index without the required probabilistic transformation. Do not assume the input means always map to the standard-space origin for arbitrary distributions. The nearest boundary point is a density-based geometric concept in transformed space, not a guarantee that one failure mechanism dominates the whole integral. Multiple design points, discontinuities and strong curvature can undermine a single tangent approximation. An optimizer reporting success does not prove the global nearest point was found. FORM sensitivities also depend on definitions and transformations and should not be treated as universal importance rankings.

Business use

Businesses commissioning simulation-based reliability studies should request the exact event definition, input model, transformation and numerical diagnostics. Use a transparent benchmark and document independent checks appropriate to the boundary's complexity. Keep uncertainty about distributions and physical modeling separate from numerical integration error. FORM can reduce computational burden and guide further analysis, but a small estimated probability may be highly sensitive to unsupported tail assumptions. Review the consequences of model error before using an approximate probability in a consequential design decision.

Check your understanding

Why is FORM exact in the worked example?

Answer: The event boundary is already a hyperplane in independent standard normal coordinates, so the tangent approximation introduces no boundary error.

Related terms

Importance sampling
Fast probability integration
reliability index
design point
limit-state function
standard normal transformation
second-order reliability method

Learn next

compare FORM with a second-order boundary approximation, then examine a problem with multiple failure regions and test how starting points affect the design-point search.

Sources

Sources: OpenTURNS, FORM documentation, design point and linearization: https://openturns.github.io/openturns/latest/user_manual/_generated/openturns.FORM.html ; OpenTURNS, illustrated FORM probability estimate: https://openturns.github.io/openturns/1.20/auto_reliability_sensitivity/reliability/plot_form_explained.html ; OpenTURNS developer discussion of failure-side orientation: https://openturns.discourse.group/t/improving-the-form-example/97 ; NASA, uncertainty propagation methods and FORM distinction: https://ntrs.nasa.gov/api/citations/20130011505/downloads/20130011505.pdf .

Previous: First-order Markov process Next: First-order second-moment method