eduKateSG · How Science Works
Choose your reading route
Follow standardised supernova light into distance–redshift tests, calibration limits and current dark-energy comparisons.
Define acceleration and measure distance · Chapters 1–4 · Start this route
Compare histories and quantify uncertainty · Chapters 5–7 · Start this route
Combine evidence and assess evolving-dark-energy claims · Chapters 8–10 · Start this route
Full contents · Check your understanding · How Science Works master guide
A distant supernova arrives as a changing point of light and a spectrum. Neither measurement comes with an acceleration label. Scientists must infer a distance, measure a redshift, compare many objects and ask which expansion histories explain the resulting pattern.
The evidence for accelerated expansion and the physical explanation called dark energy are related questions, but they are not identical. This article follows the distance–redshift test, its assumptions and the distinction between established broad behaviour and current questions about dark energy’s evolution. The Astrophysics guide provides the wider context of cosmic signals and physical inference.
Contents
Define acceleration and measure distance · Chapters 1–4
1. Acceleration refers to the scale factor
2. Redshift tells us about stretching, not acceleration by itself
3. Type Ia supernovae are standardised, not perfectly identical
Compare histories and quantify uncertainty · Chapters 5–7
5. Compare two fully specified expansion histories
6. A physical explanation adds assumptions about gravity and pressure
Combine evidence and assess evolving-dark-energy claims · Chapters 8–10
8. Cosmic rulers provide a complementary route
9. Testing evolution is more demanding than testing acceleration
CHAPTER 1 OF 10 · Define acceleration and measure distance
1. Acceleration refers to the scale factor
Cosmology describes large-scale expansion using a scale factor a(t). For ideal comoving locations, their physical separation is a(t) multiplied by a fixed comoving separation. Expansion means the scale factor is increasing. Accelerated expansion means its rate of increase is itself increasing: its second time derivative is positive.
This statement concerns the large-scale cosmological description. It does not imply that atoms, people, the Solar System and every gravitationally bound system expand in step with a. Local binding and dynamics matter.
The Hubble parameter H is the fractional expansion rate, H = ȧ/a. It is possible for ȧ to increase while H decreases, because the denominator also grows. Confusing these two rates leads to a common but important mistake.
An original mathematical example makes the distinction explicit. Let a = (t/t₀)² for positive t, where t₀ is a fixed time. Its first derivative is 2t/t₀² and its second derivative is 2/t₀², so this model accelerates. Yet H = 2/t, which decreases as time passes. At t = t₀, H = 2/t₀; at t = 2t₀, H = 1/t₀.
This power-law example is not a fitted history of our universe. It is a counterexample to the claim that acceleration necessarily requires an increasing Hubble parameter. Establishing what the word means comes before interpreting the observations.
CHAPTER 2 OF 10 · Define acceleration and measure distance
2. Redshift tells us about stretching, not acceleration by itself
Cosmological redshift relates the observed wavelength to its emitted value: 1 + z = λ_observed/λ_emitted. In the standard homogeneous expanding-universe framework, it also equals the ratio of the present scale factor to the scale factor at emission.
An observed redshift z = 0.5 corresponds to wavelengths stretched by a factor of 1.5. Normalising today’s scale factor to one gives an emission scale factor of 2/3. That identifies an expansion ratio, not the elapsed time or the second derivative of the scale factor.
Many expansion histories can reach the same ratio. A single redshift therefore cannot distinguish an accelerating universe from a decelerating one. Additional information must tell us how distance or another observable changes across redshift.
Nearby objects also have peculiar velocities relative to the large-scale flow. These motions can contribute appreciably to their observed redshifts. Spectral identification, reference-frame corrections and velocity modelling are part of the measurement chain, especially when nearby supernovae anchor a comparison with more distant ones.
For a diagram, draw several rising scale-factor curves that all connect a = 2/3 to a = 1. Give them different slopes and curvature. Add the caption “same scale-factor ratio, different histories”. The picture helps separate what redshift supplies from the further evidence needed to infer acceleration.
CHAPTER 3 OF 10 · Define acceleration and measure distance
3. Type Ia supernovae are standardised, not perfectly identical
Type Ia supernovae are useful because their light curves and colours contain information that helps standardise their luminosities. They are not all intrinsically identical bulbs. Classification, light-curve shape, colour, host-galaxy properties and the treatment of dust all enter the analysis.
The observer measures flux: power received per unit area. The source has a luminosity: emitted power. Luminosity distance D_L is defined so that the flux–luminosity relationship takes the form F = L/(4πD_L²), with the appropriate cosmological effects incorporated in D_L. This is not simply the present physical separation measured with a ruler.
Astronomers often express brightness logarithmically using magnitudes. The distance modulus is μ = m − M = 5 log₁₀(D_L/10 parsecs), after the relevant standardisation and corrections. Here m is apparent magnitude and M the corresponding absolute magnitude.
Relative distances can trace the shape of the distance–redshift relation without separately fixing the absolute luminosity scale and the present Hubble constant. An external calibration is needed to break that normalisation degeneracy. This is why “supernovae measure expansion history” should not be replaced casually by “every supernova alone measures H₀”.
The calibration chain also links surveys. If nearby and distant samples use different instruments or filters, their relative brightness scales must agree. A small redshift-dependent offset can imitate a change in the shape of the cosmological relation even when each survey has many precisely measured objects.
CHAPTER 4 OF 10 · Define acceleration and measure distance
4. Translate a magnitude residual into a distance change
Suppose a hypothetical standardised supernova is 0.20 magnitudes fainter than a specified reference model predicts at its measured redshift. Because larger magnitude means less received flux, the flux ratio is 10^(−0.4 × 0.20) ≈ 0.832.
If luminosity standardisation and photon propagation are correctly accounted for, the corresponding luminosity-distance ratio is 10^(0.20/5) ≈ 1.0965. The inferred distance is about 9.65% larger than the reference prediction. This numerical example is original; it is not a reported residual from a particular survey.
The word “reference” is essential. Fainter than which model, with which matter content, curvature, expansion rate and calibration? A magnitude residual has no standalone acceleration meaning until those choices are specified.
Nor is a fainter signal uniquely a larger distance. Dust, misclassification, luminosity evolution, calibration error or an incorrect selection correction can alter the observed comparison. Researchers use colours, spectra, light-curve information, survey simulations and complementary data to investigate those alternatives.
Riess and colleagues’ 1998 analysis used distant Type Ia supernovae to compare cosmological distance predictions and found evidence for accelerated expansion within the tested framework. The paper examined systematic explanations rather than treating dimness alone as proof. Its significance is the construction of a population-level distance test, not a claim that one faint object reveals the universe’s dynamics. Read the original study.
CHAPTER 5 OF 10 · Compare histories and quantify uncertainty
5. Compare two fully specified expansion histories
For optional mathematical depth, consider spatially flat models containing pressureless matter and a cosmological constant, neglecting radiation over the redshift range of this exercise. Write E(z) = H(z)/H₀ = √[Ω_m(1 + z)³ + Ω_Λ], with Ω_m + Ω_Λ = 1.
Their luminosity distance is D_L = (c/H₀)(1 + z) times the integral of 1/E(z′) from z′ = 0 to z′ = z. Distance therefore accumulates information over an interval of expansion history. It is not a direct instantaneous reading of acceleration at the source.
Assign model A Ω_m = 1 and Ω_Λ = 0. Assign model B Ω_m = 0.3 and Ω_Λ = 0.7. Use the same H₀ and source calibration for both. These are deliberately chosen teaching comparisons, not new fits to observations.
At z = 0.1, numerical integration gives D_L/(c/H₀) ≈ 0.10238 for A and 0.10748 for B. At z = 0.5 the values are 0.55051 and 0.66148. At z = 1 they are 1.17157 and 1.54285. Thus B predicts larger luminosity distances at these redshifts under the specified common normalisation.
The corresponding B-minus-A distance-modulus differences are approximately 0.105, 0.399 and 0.598 magnitudes. The change across redshift, rather than one arbitrary vertical offset, helps distinguish the histories. Real analysis fits nuisance parameters and covariance alongside cosmology; these clean curves omit that complexity so the geometric comparison is visible.
A useful residual plot puts redshift on the horizontal axis and distance-modulus difference from model A on the vertical axis. Model A is then a zero line, while model B rises through the three calculated values. Add error bars only if an actual measurement model supplies them; a smooth theoretical curve does not come with observational certainty.
CHAPTER 6 OF 10 · Compare histories and quantify uncertainty
6. A physical explanation adds assumptions about gravity and pressure
Within general relativity and a homogeneous, isotropic description, cosmic acceleration depends on both energy density and pressure. Writing ρ as mass-equivalent energy density, the acceleration equation contains the combination ρ + 3p/c². Sufficiently negative pressure can make the scale factor’s second derivative positive.
A cosmological constant has equation-of-state ratio w = p/(ρc²) = −1. Pressureless matter has w approximately zero. In a flat matter-plus-constant model, the present deceleration parameter is q₀ = Ω_m/2 − Ω_Λ, where q = −a ä/ȧ². Negative q means acceleration.
Our model A has q₀ = 0.5 and decelerates. Model B has q₀ = 0.15 − 0.7 = −0.55 and accelerates today. The signs follow from the assumptions, not from the labels attached to the curves.
Model B did not accelerate throughout its entire history. Matter density grows backward in time as (1 + z)³, while the cosmological-constant density stays constant. Equating their opposing contributions gives a transition redshift [2Ω_Λ/Ω_m]^(1/3) − 1 ≈ 0.671 in this simplified model.
The observation and the explanation remain distinct. A distance–redshift pattern constrains expansion history. Calling its cause a cosmological constant, a dynamical field or modified gravity introduces further predictions and tests. “Negative pressure” is also a relativistic property in the equations, not ordinary suction pulling galaxies toward some external empty region.
CHAPTER 7 OF 10 · Compare histories and quantify uncertainty
7. More objects do not erase a shared calibration error
Suppose each of 100 independent standardised supernova residuals has random scatter 0.10 magnitudes. The standard error of their mean would be 0.10/√100 = 0.010 magnitudes under the simple identical-distribution assumptions.
Now add a shared calibration uncertainty of 0.020 magnitudes, independent of the random component. The total uncertainty on the mean is √(0.010² + 0.020²) ≈ 0.0224 magnitudes. With 10,000 objects, the random part falls to 0.001, but the combined uncertainty remains about 0.0200. A common offset does not average away by adding objects affected by it.
A 0.020-magnitude shift corresponds to a distance factor 10^(0.020/5) ≈ 1.00925, about 0.925%. Whether that matters depends on the scientific question and how the offset varies across the sample. A uniform normalisation change and a redshift-dependent change need not affect cosmological parameters in the same way.
Selection provides another challenge. Near a survey’s detection threshold, intrinsically brighter or favourably scattered objects may be more likely to enter the sample. The population observed is therefore filtered by discovery, follow-up, classification and quality cuts. Simulations must reproduce that selection before estimating a correction.
The Pantheon+ cosmological analysis explicitly developed distance constraints with improved handling of systematic uncertainties across its supernova compilation. Its results depend on the cosmological model and on whether external calibration or other cosmological data are included. This is why sample size alone is an inadequate description of an expansion measurement. Read the primary analysis.
CHAPTER 8 OF 10 · Combine evidence and assess evolving-dark-energy claims
8. Cosmic rulers provide a complementary route
Baryon acoustic oscillations, or BAO, leave a characteristic statistical separation in large-scale matter clustering. Measuring the feature across the sky and along the redshift direction constrains combinations of transverse distance, expansion rate and the sound-horizon ruler scale.
The measurement uses a population’s correlation pattern, not one pair of galaxies declared to be exactly one standard ruler apart. Its conversion to absolute distances also depends on how the ruler is calibrated. Combining BAO with early-universe information adds power while importing the assumptions needed for that calibration.
DESI’s DR2 BAO analysis reported that the data alone were well described by flat ΛCDM, while some combinations with cosmic microwave background and supernova information preferred a time-varying dark-energy parameterisation. The preference depended on the data combination. That distinction is central: a joint tension is not equivalent to a single instrument directly observing a new dark-energy particle. Read the original DR2 analysis.
A July 2026 DESI manuscript added geometric information from the full shape of Lyman-alpha forest correlations. Its joint analyses again reported a preference for evolving dark energy under a specified parameterisation and stated external datasets. The additional measurement provides another constraint on expansion history; it does not uniquely establish a microscopic mechanism. Read the current primary manuscript.
Combining datasets also requires their cross-covariance when they overlap. Counting correlated measurements as independent would exaggerate the evidence. The aim is complementary information with a defensible joint uncertainty, not the largest possible list of agreeing headlines.
CHAPTER 9 OF 10 · Combine evidence and assess evolving-dark-energy claims
9. Testing evolution is more demanding than testing acceleration
A cosmological constant is a particular model. A common phenomenological alternative writes w(a) = w₀ + w_a(1 − a). Setting w₀ = −1 and w_a = 0 recovers the constant case. Allowing both parameters to vary gives the fit more freedom, but does not automatically identify a physical field responsible for that behaviour.
A better fit needs to be weighed against that additional freedom and against systematic uncertainty. Frequentist significance, Bayesian model odds and a posterior interval answer related but different questions. None should be casually translated into the probability that a particular physical mechanism is true.
The DES-Dovekie reanalysis, published in 2026, updated supernova cross-calibration, light-curve modelling and aspects of the uncertainty treatment. In its stated combined analysis, the significance against a cosmological constant decreased relative to the earlier DES-SN5YR analysis, while a preference for evolution remained. This is a concrete demonstration that improved calibration can change the strength of a subtle cosmological claim. It is not a reversal of the entire distance-based case for acceleration. Read the primary reanalysis.
As of the sources checked in October 2026, these studies motivate continued tests of evolving dark energy rather than a claim that the constant-versus-evolving question is conclusively settled. Different probes, sample choices and analysis assumptions can change the strength of the preference.
The right response is neither to declare every tension a discovery nor to dismiss every tension as an error. Specify the competing models, improve calibration, test withheld predictions and compare genuinely complementary observations. The uncertainty is part of the scientific result.
CHAPTER 10 OF 10 · Combine evidence and assess evolving-dark-energy claims
10. Audit an acceleration claim from signal to conclusion
Begin with the observables: spectra, redshifts, calibrated fluxes and light curves. Follow the standardisation and selection corrections into distances. Identify the cosmological relation being fitted, its nuisance parameters, the covariance and any external information. Finally ask whether the conclusion concerns acceleration, a cosmological constant or evidence for a more specific evolving model.
Can expansion accelerate while H decreases? Yes: a = (t/t₀)² gives positive ä but H = 2/t. Does redshift 0.5 alone prove acceleration? No; it supplies a wavelength and scale-factor ratio in the stated framework. What does a 0.20-magnitude faintness residual imply if the luminosity and propagation assumptions hold? A distance about 9.65% greater than the specified reference prediction.
Why compare a range of redshifts? Different histories change the shape of the distance relation, not just its overall normalisation. Why do thousands of measurements not erase one shared calibration offset? The same systematic contribution remains in their mean. Does evidence favouring w₀ and w_a away from the constant case identify a new particle? No; the parameterisation describes behaviour rather than a unique cause.
All numerical examples here are original calculations in explicitly simplified models. They can be explored on paper or in a spreadsheet without treating them as a new cosmological dataset. Keep assumptions beside the numbers, especially the common H₀, spatial flatness and neglected radiation in the two-model comparison.
Continue with Astrophysics or the How Science Works master guide. The lasting lesson is how scientists reconstruct a changing universe from calibrated comparisons, while keeping the observed history and its physical explanation open to distinct tests.