eduKateSG · How Science Works
Choose your reading route
Read relic radiation through temperature units, foreground subtraction, acoustic scales and model-dependent inference.
Understand the radiation and clean the map · Chapters 1–4 · Start this route
Read angular patterns and acoustic evidence · Chapters 5–7 · Start this route
Check independent data and uncertainty · Chapters 8–10 · Start this route
Full contents · Check your understanding · How Science Works master guide
The cosmic microwave background fills the sky with radiation that is remarkably close to a thermal spectrum. Across that nearly uniform background are tiny directional differences. Their pattern contains evidence about an early universe in which radiation and ordinary matter interacted much more strongly than they do across most of space today.
A colourful cosmic microwave background map is a carefully processed scientific product. Understanding it means distinguishing the average radiation temperature, small angular variations, foreground emission and the cosmological model used to interpret them. This article follows that evidence chain rather than attempting a complete history of the universe. For the wider framework, begin with Astrophysics.
Contents
Understand the radiation and clean the map · Chapters 1–4
1. The background is older than the stars we normally photograph
2. Expansion stretches the spectrum while preserving its thermal form
Read angular patterns and acoustic evidence · Chapters 5–7
5. An angular power spectrum asks which scales carry variation
Check independent data and uncertainty · Chapters 8–10
8. Independent maps can expose shared and unshared errors
CHAPTER 1 OF 10 · Understand the radiation and clean the map
1. The background is older than the stars we normally photograph
In the hot early universe, many electrons were free rather than bound into neutral atoms. Photons scattered repeatedly from those electrons. Ordinary matter and radiation were closely coupled, so a photon could not simply preserve a clear image over a very long unobstructed path.
As the universe expanded and cooled, electrons increasingly combined with nuclei. The falling free-electron abundance made the universe much more transparent. The photons we observe as the cosmic microwave background largely last scattered during this transition, about 380,000 years after the beginning of the standard hot-big-bang expansion history.
“Last scattering” is statistical. It occurred over a finite interval, not on a perfectly sharp universal shutter click. Some photons were scattered again later, particularly after stars and galaxies helped reionise gas. Other interactions and gravitational effects also modified the signal on its journey.
The sky therefore shows us a distant last-scattering region in every direction. We occupy the centre of our observable sphere because observations are made from here, not because the early universe exploded outward from our location into pre-existing empty space.
Nor is this radiation a photograph of time zero. Earlier conditions are inferred through their consequences for the radiation and matter that became observable later. Keeping that distinction clear prevents a powerful fossil record from becoming a claim to have directly watched the universe’s beginning.
CHAPTER 2 OF 10 · Understand the radiation and clean the map
2. Expansion stretches the spectrum while preserving its thermal form
In the standard expanding-universe description, freely travelling photon wavelengths stretch in proportion to the scale factor a. A blackbody radiation distribution correspondingly cools as T is proportional to 1/a, provided the evolution is adiabatic and processes that would appreciably distort the spectrum are absent.
Redshift z expresses the wavelength stretch as 1 + z. Using the approximate present temperature 2.7255 K, radiation observed from an illustrative redshift z = 1099 corresponds to T = 2.7255 × 1100 ≈ 2,998 K. This calculation connects two radiation temperatures. It does not claim that every part of the universe today is at 2.7255 K.
A wavelength example makes the same point. A photon that had wavelength 1 micrometre at that redshift would now have wavelength 1.1 millimetres. The wavelength grows by 1100, while its energy, proportional to inverse wavelength, falls by the same factor. The example follows one chosen wavelength, not a uniquely defined peak of the entire spectrum.
Measurements of the mean spectrum supply evidence separate from angular maps. Fixsen’s 2009 analysis recalibrated COBE/FIRAS using WMAP information and combined available temperature measurements, obtaining 2.72548 ± 0.00057 K. The result demonstrates both the precision of the thermal measurement and the importance of calibration. Our worked examples round the temperature rather than claiming that precision for every calculation. Read the original temperature study.
CHAPTER 3 OF 10 · Understand the radiation and clean the map
3. The colours show tiny departures from a reference
Imagine a sky direction assigned a temperature departure of +50 microkelvin, or +50 μK. Relative to 2.7255 K, that is 50 × 10⁻⁶ / 2.7255 ≈ 0.00001835: about 18 parts per million. An intensely red patch on a displayed map may therefore represent a very small excess, not an incandescent cloud visibly glowing red.
The colour scale is deliberately expanded to reveal differences that would otherwise be hard to see. The assigned hues are a visual key, not visible-light colours measured by a camera. A scientifically useful caption must state which quantity is mapped, its units and the processing applied.
There are several layers to remove or model. The nearly uniform mean is the monopole. A much larger broad directional pattern, dominated by our motion relative to the radiation rest frame, is the dipole. Cosmological temperature-anisotropy maps commonly present the much smaller structure after these large contributions have been treated.
For a schematic diagram, draw three panels with clearly different scales. Panel one is an almost uniform sphere labelled mean temperature. Panel two has a broad two-sided gradient labelled motion dipole. Panel three has smaller patches labelled residual anisotropy after processing. Do not use one colour bar for all three without explaining the enormous contrast in amplitude.
A hot-looking patch is not simply a direct density reading. Intrinsic temperature, motion of the emitting material, gravity and effects along the light path can all contribute to the observed temperature departure. The map is informative because physical models predict their combined pattern.
CHAPTER 4 OF 10 · Understand the radiation and clean the map
4. Several frequencies help separate the sky’s ingredients
Our Galaxy emits microwave radiation through several processes, including dust emission and radiation from charged particles. Distant galaxies and galaxy clusters also contribute. Instruments add noise, beam smoothing and calibration uncertainty. The desired cosmic signal is therefore mixed with other signals rather than lying alone in every detector sample.
Frequency coverage helps because these components have different spectral behaviours. Expressed in thermodynamic CMB temperature units, a small primary CMB temperature fluctuation has the same ideal amplitude across frequency channels. Foregrounds generally do not. This statement depends on the unit conversion; raw intensity is not frequency-independent.
Consider an original two-channel model in those temperature units. Let measured values be y₁ = s + d and y₂ = s + 2d, where s is the desired CMB fluctuation and d is a foreground amplitude. If the channels report 70 and 90 μK, subtraction gives d = 20 μK and then s = 50 μK.
Suppose each channel has independent noise with standard deviation 3 μK and the mixing coefficients are exact. The cleaned estimate s = 2y₁ − y₂ has noise standard deviation √(4 × 9 + 9) = √45 ≈ 6.71 μK. Removing a contaminant can amplify noise. A clean-looking map is not automatically a low-uncertainty map.
Now suppose the true second foreground coefficient is 2.2 instead of 2. The same underlying s = 50 and d = 20 produce observations 70 and 94 μK. Applying the wrong cleaning equation returns 46 μK. More channels, foreground models, sky masks and cross-checks are needed in real work; two numbers cannot identify an arbitrary collection of components.
CHAPTER 5 OF 10 · Read angular patterns and acoustic evidence
5. An angular power spectrum asks which scales carry variation
A sky map records where the fluctuations occur. An angular power spectrum summarises how much variation appears at different angular scales. The sphere is decomposed into patterns called spherical harmonics, labelled by an angular multipole l. Low l describes broad structure; high l describes finer structure.
A rough guide is angular scale in degrees ≈ 180/l. Thus l = 200 corresponds to about 0.9 degrees and l = 1000 to about 0.18 degrees. This is an orientation rule, not a statement that each multipole corresponds to a collection of identical circular spots.
The spectrum is often displayed using a rescaled quantity D_l = l(l + 1)C_l/(2π), rather than C_l itself. A figure caption matters because the two vertical axes have different normalisations. Temperature power has squared-temperature units, commonly μK²; its square root is an amplitude scale, not another power.
Different maps can share the same power spectrum while placing their hot and cold regions differently. The spectrum does not preserve every phase relationship. Its usefulness rests partly on the approximate statistical properties of the fluctuations, and researchers also study statistics beyond the ordinary two-point power spectrum.
This distinction explains why a scientific model is usually not asked to reproduce the precise location of every displayed patch. Instead it predicts statistical patterns whose comparison with the observed sky includes sampling uncertainty. One sky is a realisation of those patterns, not an unlimited collection of independently repeated universes.
CHAPTER 6 OF 10 · Read angular patterns and acoustic evidence
6. Acoustic structure remembers a coupled plasma
Before last scattering, gravity tended to draw matter into overdense regions while radiation pressure resisted compression in the coupled photon–baryon fluid. Disturbances could oscillate. Here baryons refer to ordinary matter associated mainly with protons and neutrons; electrons were essential to the coupling through scattering.
Different spatial wavelengths reached different phases of compression and rarefaction by the time the radiation decoupled. This produces a sequence of acoustic features in the angular spectra. The peaks are not a set of modern stars and galaxies already shining at that epoch, nor sound waves that travelled through today’s empty space to the telescope.
The sound horizon gives a characteristic distance that these disturbances could have travelled before last scattering. Its angular size depends on both early-universe physics and the distance geometry between us and the last-scattering region. A small-angle relation illustrates the combination: θ ≈ r/D when the ruler length r and distance D use a consistent convention.
Assign a comoving ruler of 150 megaparsecs and a comoving transverse distance of 15,000 megaparsecs. The angle is 0.010 radians, about 0.573 degrees. A ruler of 165 megaparsecs and distance 16,500 megaparsecs gives exactly the same angle. An angle alone therefore fixes a ratio, not both lengths separately.
This toy ruler is not a calculation of the observed first peak position. Driving, projection and phase effects enter the real spectrum. Its purpose is to show why an apparently simple angular measurement constrains combinations of physical parameters and must be analysed with the rest of the evidence.
CHAPTER 7 OF 10 · Read angular patterns and acoustic evidence
7. The peaks test several ingredients together
Changing the amount of ordinary matter affects the inertia of the coupled fluid. Changing the dark-matter content affects gravitational driving and the timing of matter–radiation equality. Expansion history changes the projection of physical scales onto the sky. The initial fluctuation spectrum also shapes the observed pattern.
Consequently, a cosmological fit compares many measured features at once. It cannot freely adjust one ingredient to fix one peak without checking what that adjustment does elsewhere. Polarisation and the cross-correlation between temperature and polarisation add further constraints because they respond to related physics in different ways.
The Planck team’s cosmological-parameter analysis showed how its temperature, polarisation and lensing information constrain a specified cosmological model and extensions. The familiar parameter estimates belong to stated data combinations and assumptions, including the baseline ΛCDM framework. They are not quantities printed directly on the raw microwave sky. Read the primary Planck parameter paper.
A productive reading question is: which quantities were allowed to vary? If spatial curvature, the neutrino sector or the dark-energy description changes, some inferred intervals can change too. This does not make model fitting arbitrary. Alternatives remain constrained by the full data and must make their own quantitative predictions.
Likewise, the success of a broad early-universe framework does not uniquely identify every proposed mechanism within it. CMB measurements test predictions relevant to inflation, but a matching spectrum is not a direct movie of one particular inflationary field or its entire history.
CHAPTER 8 OF 10 · Check independent data and uncertainty
8. Independent maps can expose shared and unshared errors
Repeated measurement matters even when an earlier experiment is already precise. Telescopes on the ground and satellites have different observing environments, scanning patterns and instrumental challenges. Agreement across appropriately overlapping angular scales can strengthen confidence; disagreement can reveal foreground, calibration or modelling problems.
The Atacama Cosmology Telescope’s 2025 DR6 power-spectrum analysis used temperature and polarisation observations in several frequency bands and explicitly fitted CMB plus foreground contributions. It compared parameter constraints from different combinations, including ACT with WMAP and comparisons with Planck. This is an example of testing consistency through separate data routes, rather than treating one final sky image as the only evidence. Read the original ACT analysis.
“Independent” still needs qualification. Two analyses may share calibration references, foreground assumptions, sky coverage or theoretical predictions. Even entirely separate instruments observe some of the same cosmic fluctuations. Their agreement is valuable, but one should not count every shared element as a fresh independent confirmation.
A practical check is to split observations by detector, time or frequency and compare the resulting maps. A difference map should suppress the common sky signal while revealing some kinds of instrumental inconsistency. Cross-spectra between suitable splits can avoid particular noise biases, although correlated noise and foregrounds require additional treatment.
No single null test certifies an experiment. A convincing analysis assembles a collection of checks targeted at specific failure modes, then propagates the remaining uncertainty into the cosmological conclusion.
CHAPTER 9 OF 10 · Check independent data and uncertainty
9. Some uncertainty remains even with an ideal instrument
Suppose an ideal experiment could observe the whole sky with no instrumental noise. It would still have only a finite number of independent harmonic coefficients at each multipole. The resulting statistical limitation is called cosmic variance.
For a statistically isotropic Gaussian field and the usual full-sky power-spectrum estimator, the fractional standard deviation is √[2/(2l + 1)]. At l = 2 this is about 0.632, or 63.2%. At l = 200 it is about 0.0706, or 7.06%. These are theoretical ideal-estimator uncertainties, not error bars to paste indiscriminately onto a masked, binned, noisy measurement.
Observing less sky and dealing with foreground masks complicates the covariance. Instrumental beams suppress fine-scale power; noise and cleaning uncertainty add other limitations. Binning can reduce uncertainty by combining information, while also discarding some angular resolution in the plotted spectrum.
The large fractional uncertainty at low l helps explain why a visually striking broad feature needs careful statistical assessment. If researchers notice an unusual pattern after examining many possible patterns, its apparent rarity must account for that selection. A feature chosen after looking is not equivalent to a prediction specified in advance.
Cosmic variance is not an excuse to stop testing. It is a statement about how much information a particular statistic can contain. Better observations, polarisation, complementary tracers and carefully defined hypotheses can still improve what we learn.
CHAPTER 10 OF 10 · Check independent data and uncertainty
10. Reconstruct the evidence before accepting the headline
Begin with the spectrum: is the signal consistent with relic thermal radiation? Then inspect the maps: what units, monopole and dipole treatment, foreground removal, masks and beam corrections were used? Next read the angular statistics and their uncertainties. Only then connect them to a stated cosmological model.
Does a red patch mean a hot visible-light object? No. It normally marks a small temperature departure using assigned colours. Does a cleaner map necessarily have smaller noise? No. Our two-channel subtraction increased the noise from 3 to 6.71 μK. Does measuring a ruler’s angle separately determine its length and distance? No; our two different ruler–distance pairs gave the same angle.
Why is the background not a photograph of time zero? Most of its photons last scattered much later, during the transition toward transparency. Would a perfect telescope eliminate every uncertainty? No. Finite-sky statistical information remains, alongside the assumptions needed for interpretation.
The numerical examples in this article are original constructions, not Planck or ACT measurements. A useful paper-only exercise is to change the foreground coefficient in chapter four and see how the recovered temperature changes. Keep instrument uncertainty and model mismatch in separate columns: they answer different questions.
Continue through Astrophysics or return to the How Science Works master guide. Ancient light becomes evidence through a disciplined chain of calibration, separation, statistical comparison and physical modelling. Understanding that chain makes the colourful map more meaningful, not less remarkable.