research32 min studyLesson 600 in Physics

Reaction-rate uncertainty and abundance sensitivity

Propagate correlated rate errors through a specified abundance model and separate nuclear uncertainty from sampling noise and trajectory uncertainty.

Before you begin

On this page

The question for today

What you’ll understand

Define a local logarithmic abundance sensitivity with fixed modeling assumptions.
Propagate a covariance matrix rather than assuming all rate errors are independent.
Interpret Monte Carlo ensembles, key-rate correlations, and nonlinear limitations without treating one model result as universal.

A nuclear rate can be uncertain without its final abundance being equally uncertain. Competing production and destruction routes can amplify, suppress, or cancel its influence. Sensitivity therefore belongs to a specified network, trajectory, and output time, rather than to the rate in isolation. When several nuclear inputs share measurements or theoretical assumptions, their uncertainties can be correlated. Correct propagation must preserve those relationships as well as the physical connection between forward and reverse reactions.

Define the perturbation before interpreting its sensitivity

Let an abundance be predicted by a network with a fixed temperature-density history, initial composition, and selected final time. Here a varied rate means an input thermal reactivity or decay hazard, rather than an abundance-dependent reaction flow. A local logarithmic sensitivity measures the fractional abundance response to a fractional input-rate change near the baseline. It can be positive for a production route or negative for a destruction route, and it can change across different conditions or parameter ranges. A derivative for one zone is not automatically the derivative of a mass-weighted ejecta yield containing many zones.

Finite rate variations estimate the derivative only when the response is sufficiently smooth and close to linear over the chosen interval. A large perturbation can shift the bottleneck, alter branch topology, or move freezeout, making the local slope a poor global summary. If nuclear heating changes the temperature history, a postprocessed calculation with that history held fixed can also miss feedback. Reports should therefore say what was held fixed and whether the uncertainty concerns local abundances, integrated yields, or inferred astrophysical parameters.

S_i=\left.\frac{\partial\ln Y}{\partial\ln r_i}\right|_{\rm baseline},\qquad \delta\ln Y\simeq\sum_iS_i\,\delta\ln r_i

The sensitivity and logarithmic variations are dimensionless. Log rates mean logarithms of ratios to fixed reference rates with the same units. The derivative refers to the stated network, thermodynamic trajectory, initial composition, and evaluation time; it need not remain constant for large changes.↔ Scroll wide equations horizontally.

Covariance can reinforce or cancel an abundance response

A variance estimate follows by multiplying the sensitivity vector on both sides of the covariance matrix of logarithmic rates. Off-diagonal terms matter: positively correlated changes in two production rates may reinforce, while a correlated production and destruction change can cancel. Dropping correlations is not a conservative procedure in general, because it can either enlarge or reduce the propagated variance. Common measurements, resonance energies, normalizations, and theoretical parameters are all possible sources of dependence.

Rate distributions are often positive and skewed. A lognormal approximation can be useful when multiplicative inputs dominate, but it is not guaranteed for upper limits, multimodal resonance possibilities, or sums of distinct contributions. Sampling underlying nuclear parameters can preserve temperature correlations more faithfully than drawing unrelated rates at each temperature. Forward and reverse stellar rates linked by detailed balance should also be varied consistently; independent changes can manufacture an unphysical equilibrium shift and falsely inflate an abundance sensitivity.

\sigma^2_{\ln Y}\simeq\sum_{ij}S_iC_{ij}S_j,\qquad C_{ij}=\operatorname{Cov}(\ln r_i,\ln r_j),\qquad C_{12}=\rho_{12}\sigma_1\sigma_2

C is a positive-semidefinite covariance matrix, sigma_i is the standard deviation of log rate i, and rho12 is a correlation coefficient. This local propagation assumes finite covariance and a sufficiently linear logarithmic response.↔ Scroll wide equations horizontally.

An ensemble estimates a conditional distribution

A Monte Carlo study draws uncertain inputs, evolves the network for each draw, and summarizes the resulting abundance distribution. Quantiles can remain informative when the distribution is asymmetric. The ensemble does not eliminate uncertainty in the adopted stellar model, seed composition, or trajectory unless those inputs were included in the sampling. Increasing the number of draws reduces sampling noise in estimated statistics; it does not shrink the physical uncertainty distribution assigned to the nuclear inputs.

Correlations between rate variation and final abundance can identify influential reactions in a chosen ensemble, but correlation alone is not a universal causal ranking. Shared inputs and compensation among several reactions can complicate interpretation. A large nominal reaction flow can be unimportant to a particular output if a fast reverse process cancels it. Conversely, a small branch can control a rare product. Experimental priorities should combine sensitivity, credible rate uncertainty, model robustness, and whether a proposed measurement actually constrains the relevant stellar contribution.

f=\frac{\lambda_c}{\lambda_c+\lambda_\beta},\qquad \frac{\partial\ln f}{\partial\ln\lambda_c}=1-f,\qquad \frac{\partial\ln f}{\partial\ln\lambda_\beta}=-(1-f)

This exact example is an eventual two-exit branch with constant capture and beta hazards. It shows opposite sensitivities, while a finite exposure would also depend on the total waiting time and survival probability.↔ Scroll wide equations horizontally.

Work it through

A production–destruction cancellation in log space

In a stipulated local abundance model, $S_1=1/2$ and $S_2=-1$ . The logarithmic rate standard deviations are $\sigma_1=0.2$ and $\sigma_2=0.1$ . Compute the propagated standard deviation first for independent errors and then for perfectly positively correlated log errors with these same standard deviations.

For independent errors the variance is $(1/2)^2(0.2)^2+(-1)^2(0.1)^2=0.01+0.01=0.02$ . Its square root is $\sqrt{0.02}\simeq0.1414$ in log abundance.
Perfect positive correlation adds covariance $C_{12}=0.2(0.1)=0.02$ . The cross term is $2(1/2)(-1)(0.02)=-0.02$ , so the first-order propagated variance becomes zero.
The cancellation corresponds to correlated shifts $\delta\ln r_2=(1/2)\delta\ln r_1$ , exactly opposing the two local responses. Zero first-order variance does not exclude higher-order effects, other uncertain inputs, or failure of the local sensitivity approximation.

The result

Independent log-rate errors give abundance standard deviation about 0.1414; the stated perfectly correlated changes cancel at first order.

The living notebook

Explore the model

Interactive lab

Interactive lab · sampling

Sampling precision is distinct from physical uncertainty

For independent fresh runs Reset and set p to 0.25 with n=20, then p to 0.25 with n=80. Predict exact sample-mean variances 0.009375 and 0.00234375, a decrease by four; the readouts round these values. The instrument displays exact binomial sample-mean probabilities and does not randomly resample. This illustrates estimation precision only; it does not draw uncertain nuclear rates or show that physical rate errors shrink when more network runs are performed.

Exact binomial massNormal interval massHighlighted outcome / mean

Mean sample proportion: 0.5
Sample-mean variance: 0.0125
Uniform CDF error bound: ≤ 0.1062

Success probability p0.5Independent samples n20

t = 0 s

Exact binomial masses for the mean of n independent Bernoulli trials, compared with normal mass between adjacent half-count boundaries; normal tails outside the finite support remain outside the bars. Playback highlights outcomes, not random samples. The Berry–Esseen bound applies to standardized CDFs, not individual bars or continuity-corrected errors. At p = 0 or 1, normal comparison and that bound are disabled. Normal-CDF evaluation uses a numerical approximation with absolute error below 7.5×10⁻⁸ (at most 1.5×10⁻⁷ per interval, before rounding). Parameter changes pause playback and recompute the model at the current elapsed time. Reset restores both parameters and returns time to zero. Playback starts paused. Adjust either parameter, then choose Play or Step.

Look forMore samples sharpen an estimated statistic; they do not remove the uncertainty distribution being sampled.

Pause & reason

Check your understanding

Interactive checks need JavaScript. You can still read the explanation.

Read the checkpoint explanation

Reciprocity relates the two directions through nuclear and thermal inputs. Consistent variations preserve that relation; the correct joint treatment depends on which underlying quantities are actually uncertain.

Your turn

Try it yourself

An eventual capture branch has $\lambda_c=3\lambda_\beta$ . Find its logarithmic sensitivities to the two rates. Then multiply both hazards by two and compare the branch fraction and mean waiting time. Explain why a finite exposure could still show a change.

Take a moment to make a prediction or work it out on paper. Open a hint when you need one.

Give me a hint

Use the branch fraction and its complementary fraction. A common rate factor cancels in a ratio but not in a lifetime.

Compare with the solution

The branch fraction is $f=3/4$ , giving sensitivities $+1/4$ and $-1/4$ . Doubling both rates preserves $f=3/4$ , while the mean waiting time $1/(\lambda_c+\lambda_\beta)$ halves. During a fixed finite exposure more parents therefore reach either exit, even though the conditional eventual split is unchanged. Sensitivity depends on which observable and time interval are specified.

Keep the idea

Take it with you

Rate sensitivity is conditional on the network, trajectory, and chosen output.
Correlations and detailed-balance links belong in uncertainty propagation.
Monte Carlo sampling precision differs from nuclear-input and astrophysical-model uncertainty.

Read further

Original sources and deeper explanations to follow at your own pace.

Optional progress, saved in this browser when available. Revisit any lesson at any time.

Lesson not marked complete.