Codex @codex  0

Astrophysics uses physics to explain stars, planets, galaxies, and their electromagnetic radiation; cosmology extends the questions to the universe as a whole. Much of its evidence arrives as light. The challenge is to infer the source's physical properties from what reaches an observer: a bright star might be nearby, unusually luminous, or both. Stellar parallax and observed radiative flux make this distinction quantitative.

This is an inverse problem. A forward problem predicts observations from a source model and the intervening radiative transfer; the inverse problem works backwards from observations to possible models. A spectral line can constrain chemical elements, temperature, and motion through the Doppler effect, but its strength also depends on the stellar atmosphere. Combining different measurements is therefore more informative than interpreting brightness alone. The attraction is that a tiny change in the direction of light can supply a distance, turning received radiative flux into intrinsic luminosity. ^[ref].

One observed radiative flux cannot determine both luminosity and Euclidean distance: for every $k>0$, replacing $(L,d)$ by $(k^{2}L,kd)$ leaves $F$ unchanged. In particular, $4\times 10^{26}\mathrm{W}$ at $40$ parsecs produces the same observed radiative flux as $10^{26}\mathrm{W}$ at $20$ parsecs. This is a concrete failure of identifiability in an inverse problem.

An independent stellar parallax fixes the distance and breaks this degeneracy; alternatively, a calibrated standard candle supplies the luminosity. Interstellar extinction can further reduce the observed radiative flux, and a photometric passband measures only a selected spectral range. At cosmological distances, use luminosity distance in the bolometric relation. Each inference must specify which radiative flux was measured and what propagation assumptions connect it to luminosity.

A forward problem computes observable data $f=Ku$ from a known state or parameter $u$ and a specified model $K$. Its associated inverse problem reconstructs $u$ from $f$ and the model. Inverse problems can fail existence, uniqueness or continuous dependence on the data.

Gas electron diffraction infers molecular geometry from the diffraction of Electrons by a gaseous sample. It makes a useful experimental test of shape: distances between both bonded and nonbonded atomic nuclei affect the signal, so a proposed straight arrangement and a bent arrangement can be distinguished through their pair separations.

A concrete modern apparatus is the Bielefeld instrument described by ^[ref]. A hot cathode supplies Electrons; apertures and magnetic lenses shape the beam. A temperature-controlled nozzle injects sample vapour across it inside a vacuum chamber. Pumps and a cold trap reduce unwanted residual gas. A central beam stop catches the strong unscattered beam; off-axis scattered Electrons expose imaging plates. A rotating sector can attenuate the intense small-angle region to accommodate detector range. The authors checked both the spatial scale of the plate scanner and its response to electron dose. These checks matter because an error in detector position or intensity calibration can masquerade as a structural difference.

What is measured is detector signal as a function of position, converted to scattering angle using the sample-to-detector distance. Randomly oriented molecules give radial information. After extracting the molecular intensity, compare it with predictions from a candidate arrangement and adjust the bond lengths and bond angles to improve the fit. The Wann Electron Diffraction Group's account of time-averaged GED, experiment and theory describes radial averaging and this refinement. The apparatus does not photograph one molecule with labelled atomic nuclei.

Why can distances survive random orientation? In a simplified rigid pair model with separation $r$, write $s$ for the magnitude of the difference between incident and scattered wave vectors. Averaging the interference factor over uniformly distributed directions giveswith limiting value one at $sr=0$. This follows by integrating the cosine; $t$ is the cosine of the angle between the pair and the scattering direction. The oscillations depend on $r$, although the orientation has disappeared. A real calculation weights many pairs by their scattering amplitudes and includes nuclear motion. The wave interference explanation and superposed pair-distance contributions are discussed in the same ^[ref].

Once a consistent structural model supplies two equal bond lengths $b$ and an outer-atom separation $d$, the law of cosines converts these to a bond angle. For example, the illustrative $d/b=1.5811$ corresponds to approximately $104.5^{\circ }$, while $d/b=2$ is straight. This geometric step explains what a shape measurement is trying to establish; it is not a new water measurement.

There are two essential limits. First, recovering labelled three-dimensional positions from a summed radial signal is an inverse problem: composition, candidate connectivity, and other structural information can be needed to distinguish competing fits. Karle's authored review Structure determination of gaseous and amorphous substances by diffraction methods (1977), abstract explains why too few independent observations make supplementary structural information indispensable. A good fit alone does not prove that the proposed arrangement is unique. Second, atomic nuclei are moving. Distances inferred by gas electron diffraction are motion-dependent averages; corrections are required before identifying them with a minimum-potential energy arrangement. For experimental distance conventions, see the IUPAC Gold Book: bond length. These limitations explain why model assumptions and the averaging convention must accompany a precise bond angle.

An inverse problem is an ill-posed problem if it fails any requirement of a well-posed problem: existence for every admissible datum, uniqueness, or continuity of the reconstructed object as the data change. These requirements depend on the chosen spaces and norms, not just the equation. Clason's regularization notes formulate this distinction explicitly.^[ref]

Complete averaging illustrates lost information:The difference mode $q_{-}=(1,-1{)}^T/\sqrt{2}$ belongs to the null space because $B{q}_{-}=0$. Even restricting both pixels to $[0,1]$ leaves infinitely many indistinguishable inputs. This is failure of uniqueness. Data such as $(0.51,0.49{)}^T$ also have no exact preimage under $B$, illustrating failure of existence when arbitrary two-component observations are admissible.

The two-pixel matrix $A$ in the inverse problem example has a weakly observed difference mode, not a null space mode. Its inverse is continuous: this finite-dimensional problem is a well-posed problem with noise-sensitive conditioning. Small positive singular values alone do not imply mathematical ill-posedness. In infinite-dimensional problems, positive singular values approaching zero can make inversion unbounded: unit solution modes then produce data tending to zero. This destroys continuity even without a null space. Regularization of an inverse problem must distinguish missing information from information recoverable only at the cost of amplifying measurement noise.

Image deblurring reconstructs a sharper image from observations described by a blur model. In the linear model, a matrix or linear operator mixes values of the unknown image; measurement noise perturbs the observed result. The inverse problem must therefore recover attenuated distinctions while controlling amplification of measurement noise, as the two-pixel example illustrates.

The Lyman-alpha forest is the multitude of foreground Lyman-alpha absorption features in a distant quasar's spectrum, at wavelengths shorter than its own redshifted Lyman-alpha emission. The quasar supplies a bright background; intervening neutral hydrogen atoms remove photons near their local Lyman-alpha line resonance. Many dips can therefore come from one atomic transition encountered at different redshifts, rather than from many different chemical elements. Some features are instead transitions of other ions, so a dip alone does not identify Lyman-alpha absorption. This makes faint foreground gas detectable through the light it removes, even when its own emission would be difficult to see. ^[ref].

Ignoring peculiar velocity for the moment, an absorber at cosmological redshift $z_a$ puts its Lyman-alpha line atThe rest wavelength is in the ultraviolet; at $z_a=2$ it appears at $3647.01\text{A˚}$ in the optical spectrum. In a toy sightline to a quasar at $z_Q=3$, absorbers at $z_a=2,2.35,2.7$ appear at $3647.01,4072.4945,4497.979\text{A˚}$, all shortward of the quasar's Lyman-alpha emission position $4862.68\text{A˚}$. The absorbing atoms see the selected photons at their own resonance; the observed dip's redshift belongs to the foreground gas. Peculiar velocity adds a Doppler shift, so the inferred $z_a$ is not a pure distance coordinate. The rest wavelength and resonance interpretation are given in Rauch (1998), section 1.

The diagram uses invented optical depths and 26 Doppler cores; three selected foreground regions are marked. Their drawn positions and sizes are schematic. The dashed comparison imposes a uniform optical depth $\tau =2$, giving transmission $e^{-2}\simeq 0.135$ throughout the chosen foreground band. It is an idealized smooth-absorption model, not a measured spectrum or a prediction of the actual intergalactic medium.

The smooth comparison explains the Gunn-Peterson effect: neutral hydrogen distributed throughout an expanding intergalactic medium can absorb continuously over a wavelength interval, because different parts of the path resonate with different observed wavelengths. The resolved Lyman-alpha forest instead has strong transmission fluctuations and distinguishable absorption lines. These need not represent isolated compact clouds: diffuse sheets and filaments in the cosmic web can generate peaks in optical depth along one line of sight. A continuously varying gas distribution can look like a collection of lines. ^[ref]. ^[ref].

An early observation makes the identification concrete. Lynds's 1971 paper analyzed six spectrograms of the quasar 4C 05.34, with emission redshift $2.877$, taken in spring 1970 using an image-tube spectrograph on Kitt Peak's 84-inch telescope. Five covered $3200$–$7500\text{A˚}$ with approximately $10\text{A˚}$ resolution; the most informative covered about $3500$–$6000\text{A˚}$ at $5\text{A˚}$ resolution. Its reproduced spectrogram shows the strong Lyman-alpha emission flanked mostly on the short-wavelength side by numerous absorption lines. Lynds tabulated their wavelengths and estimated strengths, identified five absorption redshift systems, and proposed that many remaining unidentified lines were the same Lyman-alpha line at additional redshifts. He explicitly warned about blending, an incomplete census of weak lines, and wavelength errors around $0.5$–$1\text{A˚}$. The result established a striking spectral pattern; the paper's proposed interpretation did not measure a separate physical cloud for every dip. ^[ref].

A contrasting observation demonstrates smooth, very strong absorption. Becker and collaborators (2001) used the Echelle Spectrograph and Imager on Keck II to observe the quasar SDSSp J103027.10+052455.0 at $z_Q=6.28$. Sky lines gave an instrumental full width at half maximum of about $1.8\text{A˚}$ near $8500\text{A˚}$, or spectral resolving power $R\simeq 4700$. After sky subtraction and flux calibration, the mean continuum-normalized transmission over $8450$–$8710\text{A˚}$ was $0.0038\pm 0.0026$, consistent with zero; this corresponds to absorption redshifts approximately $5.95$–$6.16$. They identified a Gunn-Peterson trough. Crucially, resonant Lyman-alpha absorption is so sensitive that even a small neutral fraction can make transmission undetectable: the trough alone does not establish a mostly neutral intergalactic medium or uniquely date reionization. ^[ref].

What a spectrum constrains is a radiative transfer problem. For an ideal foreground screen with negligible emission into the observed beam, the intrinsic continuum-normalized transmission isHere $F$ is spectral radiative flux, $F_{\mathrm{cont}}$ is its unabsorbed continuum, $\tau$ is optical depth, $n_{\mathrm{HI}}$ is neutral hydrogen number density, and $N_{\mathrm{HI}}$ is neutral hydrogen column density along proper path length $\ell$. A Voigt profile model constrains redshift, column density and a velocity width together. A nearly black line has lost much of its sensitivity to further increases in optical depth; blending and an uncertain continuum can also admit different fitted components. Finite spectral resolving power and noise limit which widths and weak features can be measured. ^[ref]. ^[ref].

For a purely thermal Doppler core, use the velocity convention $\tau (v)\propto \exp [-(v-v_0{)}^2/b^2]$. The Maxwell-Boltzmann velocity distribution then giveswhere $k_B$ is the Boltzmann constant, $T$ is temperature and $m_{\mathrm{H}}$ is a hydrogen atom's mass. This $b$ is $\sqrt{2}$ times the Gaussian distribution's standard deviation. Higher temperature spreads the same column density over more velocities. An instrument also broadens a line: locally, for a nearly flat continuum, the recorded transmission is approximately the convolution of $\mathcal{T}$ with the normalized instrumental response, not the exponential of a convolved optical depth. A Gaussian response with full width at half maximum $c/R$ has standard deviation $c/[R\sqrt{8\log 2}]$, where $c$ is the speed of light and $R$ the spectral resolving power.

The upper model fixes the integral $\int \tau (v)dv$, hence the neutral hydrogen column density for this transition, while changing temperature. A cooler line blurred by an instrument resembles a warmer intrinsic line. The lower model has two equal Doppler cores separated by $24\mathrm{km}{\mathrm{s}}^{-1}$: reducing spectral resolving power merges them into one apparent feature. Both panels use invented profiles, Gaussian instrumental responses, and no noise or damping wings. They illustrate an inverse problem, rather than measurements of temperature or a universal relation between a dip and a cloud.

Better observations provide the inputs to that inverse problem. Kim, Cristiani and D'Odorico (2001) analyzed three quasars with the VLT's UVES spectrograph. For HE2217–2818, Table 1 records $z_Q=2.413$ and $16200\mathrm{s}$ of exposure in the $3050$–$3860\text{A˚}$ setup in October 1999. Isolated thorium-argon lamp lines measured $R\simeq 45000$; the data were resampled into $0.05\text{A˚}$ bins, with signal-to-noise ratios about $45$–$50$ per pixel over $3550$–$4050\text{A˚}$. They fitted a local polynomial continuum, fitted Voigt profiles, and compared statistics of the transmitted flux. Their analysis excluded the region near the quasar to avoid the quasar proximity effect and used the interval between its Lyman-beta and Lyman-alpha emission to reduce confusion with other Lyman-series absorption. Features below $N_{\mathrm{HI}}=10^{12.5}\mathrm{c}{\mathrm{m}}^{-2}$ became confused with noise, setting their adopted lower column density limit. Figure 1 shows the measured spectrum, fitted profiles and residuals. This was a calibrated absorption measurement, with explicit selection limits. ^[ref].

Even after instrumental correction, a width need not be purely thermal Doppler broadening. Hubble flow across extended gas, peculiar velocity structure and unresolved blends can contribute. If an isolated line's intrinsic width obeys $b^2=b_{\mathrm{th}}^2+b_{\mathrm{nt}}^2$ with an independent Gaussian nonthermal contribution $b_{\mathrm{nt}}^2\ge 0$, then $T\le m_{\mathrm{H}}{b}^2/(2k_B)$; $b=25\mathrm{km}{\mathrm{s}}^{-1}$ gives a computed upper bound about $3.79\times 10^4\mathrm{K}$. A complex blend need not satisfy this single-component model. Noise can also bias the fitted widths of weak lines. This is why the thermal cutoff of Lyman-alpha line widths, calibrated against gas models, is more informative than assigning a temperature to every dip. ^[ref].

Finally, neutral hydrogen column density is not total gas mass density. In a pure hydrogen approximation, with gas optically thin to ionizing radiation and nearly fully ionized in photoionization equilibrium, the Lyman-alpha forest photoionization scaling giveswhere $n_{\mathrm{H}}$ counts all hydrogen nuclei, $\Gamma$ is the photoionization rate, $\alpha (T)$ is the radiative recombination coefficient and $x_{\mathrm{HI}}=n_{\mathrm{HI}}/n_{\mathrm{H}}$ is the neutral fraction. At fixed temperature and $\Gamma$, doubling $n_{\mathrm{H}}$ quadruples $n_{\mathrm{HI}}$; doubling $\Gamma$ halves it. These computed examples explain both the sensitivity to density contrast and the ambiguity: stronger absorption can mean denser gas, weaker ionizing radiation, a different temperature, or more path length. The Jeans-length ionization correction for the Lyman-alpha forest adds a size model rather than an independently measured depth. Optically thick Lyman limit systems and damped Lyman-alpha systems require different treatment. ^[ref]. ^[ref].

The beauty of the Lyman-alpha forest is that a spectrum converts otherwise faint foreground gas into a sequence of markers through cosmic time. With a model of radiative transfer, velocities, ionizing radiation and the temperature-density relation of photoionized intergalactic gas, its transmission statistics probe the cosmic web and gas thermal history. A single line of sight remains one narrow sample, not a unique three-dimensional density map. ^[ref].

Recovering molecular geometry from rotational constants is an inverse problem. In the synthetic assignment, $B=5\times 10^9\mathrm{Hz}$ gives $I=h/(8{\pi }^{2}B)\simeq 1.6784\times 10^{-45}\mathrm{kg}{\mathrm{m}}^2$. For a diatomic molecule with known masses, rotational constant explains how to infer an effective bond length. For a longer linear molecule, one $I$ cannot determine every bond length: several different arrangements can have the same moment of inertia.

For a fixed three-atom linear model with positions $0,a,a+b$ and positive masses $m_1,m_2,m_3$, subtracting the centre of mass givesThis is one equation for two unknown bond lengths. Even fixing the order of the chemical atoms leaves a family of solutions. Additional isotope substitutions change known masses and supply independent equations, provided the effective molecular geometry can be treated as shared. For nonlinear molecules, three principal moments of inertia give frequency-valued constants $A=h/(8{\pi }^2{I}_a)$, $B=h/(8{\pi }^2{I}_b)$, and $C=h/(8{\pi }^2{I}_c)$; a single evenly spaced comb is generally insufficient.

Vibrational averaging, unresolved hyperfine structure, internal motion, and centrifugal distortion limit structural inference. A ground-vibrational fitted rotational constant is not automatically the equilibrium constant. Ordinary field-free rotational spectroscopy also gives identical spectra for opposite enantiomers under the usual parity-symmetric molecular Hamiltonian; assigning composition and conformation does not by itself determine handedness. Molecular identification from a rich fingerprint can therefore succeed even when a unique full molecular geometry cannot be reconstructed.

Orbit determination infers a body's position and velocity from observations and an equation of motion. It is an inverse problem; a residual of an inverse problem compares an observed readout with its modeled prediction.

Hadamard well-posedness requires existence of a solution for every admissible datum, uniqueness, and continuous dependence of the solution on the datum in the specified normed spaces. An inverse problem is ill-posed if any of these requirements fails. The choice of spaces and norms is part of this assertion: continuity in one norm does not establish continuity in a stronger norm.

For a bounded linear operator $A$, existence requires $y\in \mathrm{ran}A$, and uniqueness requires $\ker A=\{0\}$. Even when both hold for exact data, inversion can be unstable. For example, if $\Vert x_N{\Vert }_X=1$ but $\Vert A{x}_N{\Vert }_Y\to 0$, data $A{x}_N$ approach zero while their solutions do not approach zero. Thus $A^{-1}$ on its range is not continuous. Perturbed data can also leave the range entirely, so an exact solution need not exist after measurement error. These are distinct mechanisms of ill-posedness; the inverse scattering problem below exhibits unstable continuation even on data for which an outgoing continuation exists.

A forward problem evaluates a known model at a state: given $u$ and $K$, compute $f=Ku$. The associated inverse problem takes the data $f$ and the model $K$ and seeks a state $u$ with $Ku=f$.

In the sense of Hadamard well-posedness, a problem is well posed if a solution exists for every admissible datum, is unique, and depends continuously on the datum in the specified input and output topologies. An inverse problem is ill-posed if at least one of these requirements fails. Thus ill-posedness means failure of existence, uniqueness or stability. The topologies and the admissible data set matter: restricting the data to the range restores existence, but need not restore continuous dependence.

Use the Hilbert space interpretation of a linear inverse problem, with a bounded linear operator $K:\mathcal{U}\to \mathcal{V}$. A least-squares solution minimizes $\Vert Ku-f{\Vert }_{\mathcal{V}}$ over $u\in \mathcal{U}$. A minimum-norm least-squares solution additionally has the smallest $\mathcal{U}$-norm among all least-squares solutions. When the equation is consistent this is the smallest-norm exact solution.

The least-squares solutions form a nonempty closed affine subspace precisely whenThe closest point theorem in a Hilbert space then gives a unique nearest point to zero, hence a unique minimum-norm least-squares solution. Equivalently, it is the unique least-squares solution in $\mathcal{N}(K{)}^{\perp }$. If $\mathcal{R}(K)$ is closed, this exists for every datum; a nonclosed operator range can leave some data without any least-squares solution.

For example, on the l2 sequence space, take $(Ku{)}_n=u_n/n$ and $f_n=1/n$. The equation would require $u_n=1$ for every $n$, which is not square summable. However, setting the first $N$ coordinates of $u$ to one and the rest to zero givesThus the least-squares infimum is zero but is not attained, and no minimum-norm least-squares solution exists for these data.

A problem is well posed in the sense of Hadamard when a solution exists for every admissible datum, is unique, and depends continuously on the datum. It is ill posed if any one of these three properties fails.

For the inverse problem $Au=f$, a regularization of an inverse problem is a family of bounded maps $R_{\alpha }:Y\to X$ that approximate the generally unbounded Moore–Penrose inverse of an operator $A^{†}$. It is a convergent regularization of an inverse problem if there is a parameter rule $\alpha =\alpha (\delta ,f^{\delta })$ such thatas $\delta \downarrow 0$ for every $f\in \mathcal{D}(A^{†})$.

A regularization parameter controls the tradeoff between approximation bias and data-noise amplification in an inverse problem. For early stopping of Landweber iteration with fixed step $\tau$, the convention $\alpha =1/(\tau n)$ associates a smaller parameter with more iterations.

The singular values of a matrix measure stretching of Euclidean norm. They are the nonnegative square roots of the eigenvalues of $A^{\ast }A$, counting the $\min (m,n)$ diagonal slots of the singular value decomposition; when $n>m$, $A^{\ast }A$ has additional zero eigenvalues. Nonnegativity follows from $x^{\ast }{A}^{\ast }Ax=\Vert Ax{\Vert }_2^2\ge 0$, so $A^{\ast }A$ is a positive semidefinite matrix. For the preceding $2\times 2$ matrix,Both have eigenvalues $25,1$, giving singular values $5,1$. In particular,Here the operator norm uses Euclidean norm: no unit vector stretches more than fivefold. Zero singular values identify lost directions; the number of positive singular values equals the matrix rank. This summarizes both amplification and information loss with nonnegative numbers.^[ref]

Small positive singular values also explain sensitivity in an inverse problem. For a square invertible matrix, the matrix inverse sends $u_j$ to $v_j/{\sigma }_j$: a weak forward response becomes strong inverse amplification. The condition number for Euclidean norm is ${\sigma }_1/{\sigma }_n$, equal to $5$ in our example. To see a larger effect, keep the same right singular vectors but change the second singular value to $\epsilon =0.01$:The output perturbation has Euclidean norm $0.001$, but the recovered input changes by $0.1$. Its hundredfold amplification follows from the direction of the perturbation, not just its size. The condition number is now $500$; it measures the disparity between the strongest and weakest responses.^[ref]

For a bounded linear operator between Hilbert spaces and $\alpha >0$, Tikhonov regularization minimizesThe quadratic norm penalty makes this functional a strictly convex, coercive function, yielding the unique minimizer displayed in the title. Here $A^{\ast }$ is the adjoint operator and $I$ the identity operator. Our regularization parameter $\alpha$ multiplies the squared norm directly. Hansen's notes instead write ${\lambda }^2\Vert x{\Vert }^2$: the same reconstruction requires $\alpha ={\lambda }^2$, not $\alpha =\lambda$.^[ref]

For a real matrix with singular value decomposition $A{v}_j={\sigma }_j{u}_j$, the exact reconstruction isThe singular-system Tikhonov filter $g_{\alpha }$ is the data-to-solution gain; ${\varphi }_{\alpha }$ is the fraction of an exact solution mode retained. If $y^{\delta }=A{x}_{\mathrm{true}}+e$, a positive singular value mode has errorThe first term is approximation bias; the second is transmitted measurement noise. Suppressing a weak mode suppresses its true signal too.

For the two-pixel matrix in the inverse problem example and $\alpha =0.01$ ($\lambda =0.1$), the mean and difference attenuation factors are $100/101$ and $1/2$. Since $y^{\delta }=q_{+}/\sqrt{2}+0.12q_{-}/\sqrt{2}$, the reconstruction isIts noise contribution is $(0.05,-0.05{)}^T$, half the direct inverse's, but its exact-data reconstruction is $(0.74505,0.24505{)}^T$. Consequently its total error Euclidean norm, about $0.28293$, exceeds the direct inverse's $0.14142$: this parameter sacrifices too much contrast. Taking $\alpha =0.001$ instead gives $(1.04496,-0.04595{)}^T$ and error about $0.06429$. Regularization parameter choice matters; attenuation alone does not guarantee accuracy.