Concept Architecture
Concept
Theoretically, Nelder-Mead Method is a derivative-free numerical optimisation algorithm used to locate the minimum or maximum of an objective function when derivatives are unavailable, unreliable, or computationally expensive to obtain. The method belongs to the class of direct search optimisation algorithms and operates by iteratively transforming a simplex in the parameter space. It exists to solve nonlinear optimisation problems where gradient-based methods are unsuitable, a common situation in health economic model calibration and parameter estimation.
Mathematically, the Nelder-Mead method represents the optimisation problem by evaluating the objective function at the vertices of a simplex containing n + 1 points in an n-dimensional parameter space. During each iteration, the simplex is modified using reflection, expansion, contraction, or shrinkage operations according to objective function values. The algorithm progresses until convergence criteria based on simplex size or objective function improvement are satisfied.
In practice, the Nelder-Mead method is implemented by defining an objective function, selecting initial parameter estimates, constructing an initial simplex, and iteratively updating the simplex until an acceptable optimum is identified. In health economics it is widely applied to calibrating Markov models, compartmental models, transmission models, survival models, and other decision-analytic models where likelihood functions or error measures are complex and derivatives are unavailable.
Purpose
Used to estimate optimal model parameters by minimising or maximising an objective function without requiring analytical derivatives, particularly during calibration of complex health economic models.
Mathematical Formulae
Primary Formula
The optimisation problem is defined as
min?x ? ?�? f(x)
where:
- f(x) is the objective function
- x is the vector of decision variables or model parameters
Supporting Formulae
Reflection:
x? = x? + �(x? ? x?)
Expansion:
x? = x? + ?(x? ? x?)
Outside contraction:
x?c = x? + ?(x? ? x?)
Inside contraction:
x?c = x? ? ?(x? ? x?)
Shrink:
x? = x? + �(x? ? x?)
where:
- x? = centroid of the simplex excluding the worst point
- x? = worst vertex
- x? = best vertex
- � = reflection coefficient (typically 1)
- ? = expansion coefficient (typically 2)
- ? = contraction coefficient (typically 0.5)
- � = shrink coefficient (typically 0.5)
Related Mathematical Methods
- Direct search optimisation
- Simplex optimisation
- Function minimisation
- Model calibration
- Maximum likelihood estimation
- Least-squares optimisation
Example
A health economist calibrates a Markov model to observed survival data by minimising the sum of squared errors between predicted and observed survival probabilities.
Objective function:
min ?????? (S????????? ? S??????)�
The Nelder-Mead method begins with an initial simplex of transmission-rate and mortality-rate parameters. After successive reflection, contraction, and expansion operations, the objective function decreases from 18.4 to 0.62, indicating substantially improved agreement between the model and observed data.
Excel Implementation
| Function | Example Formula | Health Economics Application |
|---|---|---|
| SUMXMY2 | =SUMXMY2(B2:B11,C2:C11) | Calculates the sum of squared errors used as the objective function. |
| SUMSQ | =SUMSQ(B2:B11-C2:C11)* | Measures model fit when constructing objective functions. |
| Solver Add-in | Minimise objective cell by changing parameter cells | Performs derivative-free parameter optimisation. |
| ABS | =ABS(B2-C2) | Calculates absolute prediction errors for alternative objective functions. |
*Implemented in Excel using array calculations or equivalent dynamic array formulas.
VBA (Optional)
A VBA procedure can automate repeated Nelder-Mead optimisation runs with different starting simplexes to improve calibration robustness and assess convergence.
Sources
- Nelder JA, Mead R. A Simplex Method for Function Minimization. Computer Journal. 1965;7(4):308?313.
- Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation. Oxford University Press.
- Drummond MF, Sculpher MJ, Claxton K, Stoddart GL, Torrance GW. Methods for the Economic Evaluation of Health Care Programmes. Oxford University Press.
- NICE. Health Technology Evaluation Manual.
- ISPOR Good Practice Reports on model parameter estimation and calibration.
Related Concepts (2)
Library
Publications
1
A Taxonomy of Model Structures for Economic Evaluation of Health Technologies — Brennan, Chick & Davies, Vol. 15, No. 12 ed., 2006 (Health Economics)
An influential paper classifying decision-analytic model structures along axes of expected value vs randomness, entity heterogeneity, and Markovian vs non-Markovian structure — providing a framework for choosing between decision trees, Markov cohort models, microsimulation, discrete event simulation and system dynamics.
Journal ArticleView source →
Frequently Asked Questions (6)
What is the Nelder-Mead method?
A numerical optimisation algorithm searching for a function's minimum using a moving shape, called a simplex, without needing the function's derivative.
Source: Nelder & Mead 1965
What kind of problems suit the Nelder-Mead method?
The method suits problems where the objective can be evaluated but its slope cannot be calculated, either because no formula for the derivative exists or because the function is noisy or awkward. Because it searches using only the values of the function at a set of trial points, moving a geometric shape through the space toward lower values, it needs no gradient. This makes it a convenient choice for fitting or calibrating models whose internal workings do not yield derivatives. It trades speed for this generality. Nelder and Mead (1965) introduced it.
Source: Nelder & Mead 1965
How does the Nelder-Mead method work?
The Nelder-Mead method works by maintaining a simplex of points, evaluating the function at each, and iteratively replacing the worst point with a better one through geometric operations: reflecting it through the opposite face, expanding further if that improves, contracting toward the better points if not, and shrinking the whole simplex if needed. These moves let the simplex crawl downhill and adapt its shape to the function, converging toward a minimum using only function evaluations, without any gradient information.
Source: Nelder & Mead 1965
Why is the Nelder-Mead method derivative-free?
The Nelder-Mead method is derivative-free because it relies only on comparing the function's values at the points of the simplex, using geometric operations to move toward lower values, rather than on the function's gradient. This makes it useful when derivatives are unavailable, difficult to compute, or unreliable, such as for functions that are noisy, discontinuous, or defined only by a simulation. Being derivative-free is its main advantage, allowing it to optimise functions that gradient-based methods cannot easily handle.
Source: Nelder & Mead 1965
When is the Nelder-Mead method used?
The Nelder-Mead method is used for optimising functions of a few parameters where derivatives are unavailable or unreliable, such as fitting or calibrating models whose objective comes from a simulation, or minimising noisy or non-smooth functions. Its simplicity and derivative-free operation make it convenient for such problems. It is well suited to low-dimensional problems, but its performance can deteriorate in higher dimensions, so it is chosen where the number of parameters is modest and gradients are not readily available.
Source: Nelder & Mead 1965
What are the limitations of the Nelder-Mead method?
The Nelder-Mead method does not guarantee finding the global minimum and can converge to a local one or, in some cases, stall without reaching a true minimum. Its performance degrades as the number of parameters grows, making it less suitable for high-dimensional problems, and it can be slow. Lacking gradient information, it may also struggle on certain functions. Where derivatives are available and the function is smooth, gradient-based methods are often more efficient, so Nelder-Mead is reserved for derivative-free, low-dimensional problems.
Source: Nelder & Mead 1965
Trust Record
Verified by Dr Darrin Baines
British health economist
Professional identity: darrinbaines.org
Verification date: 2 Oct 2025
Content version: 1.0.0
Canonical Identity
- Persistent URI
- https://healtheconomics.wiki/concept/nelder-mead-method
- Term code
- HE-EM-DM-069
Stable URI · Machine-readable · Resolvable · CC BY 4.0