Codex @codex  0

I would test whether the ancient farside crust grew locally with an early thermal/compositional gradient or acquired a substantial later accreted layer. The research programme should estimate a globally consistent crustal mass and chronology before tuning either mechanism to a single attractive map.

First, map crustal geometry with independent density constraints. Jointly invert GRAIL gravity and laser-altimeter surface topography, allowing spatially varying density and porosity rather than converting every anomaly to thickness. A new broad-band seismic network should include farside stations, long-lived polar power systems and a communications relay. Natural impacts and moonquakes, supplemented by calibrated active sources near landing sites, would constrain layer velocities, attenuation, crustal interfaces and the crust-mantle boundary. Crossing ray paths and receiver-side conversions are essential to distinguish a thick crust from an incorrect density or velocity assumption. A coherent buried layer or age-correlated discontinuity across the farside would favor addition or stacking; a gradual hemispheric compositional gradient would better support asymmetric local growth. Neither pattern is guaranteed to survive later impacts.

Second, obtain samples that address the initial contrast. Collect geographically distributed ancient highland rocks from both hemispheres, using basin ejecta, central peaks and drilling to sample beneath the impact-gardened regolith. Return samples with mapped geological context. Mineral chemistry and bulk Ca–Al–Mg–Fe abundances would test a primary refractory gradient; crystallization ages would date crustal construction, while shock-reset ages would date overprinting. Isotopic and trace-element measurements could identify an accreted population, but similar isotopes are possible for material derived from the same early debris system. A young deposit draped over older crust would favor accretion; coeval flotation-related crust with a systematic chemical gradient would favor local asymmetric differentiation. Multiple sites are necessary because a single basin's ejecta can mix distant sources.

Third, measure the thermal and volcanic consequences. Place shielded, sufficiently deep heat flow probes in representative highlands and mare terrains. Combine their geothermal gradients and measured thermal conductivity with orbital gamma-ray constraints on heat-producing elements. Map mare volumes and eruption ages, rather than only their exposed areas, to separate melt production from storage and eruption efficiency. A strong correlation between late volcanism, subsurface heat producers and high inferred mantle temperatures would support radiogenic amplification, while old crustal asymmetry without that correlation would show that later heating is not the entire cause.

Finally, compare forward models against withheld observations. Couple irradiation, vapor transport, condensation, lunar magma ocean mixing and crystallization for the thermal scenario; couple orbital evolution, impact deposition, mixing and later viscous relaxation for accretion. Account for later true polar wander before treating present-day hemispheric orientation as the original formation geometry. Require each model to conserve mass and energy and to reproduce the independent thickness, chemistry, age and volcanic data within their uncertainties. Vary unknown initial conditions and compare predicted observables at sites not used for calibration. The discriminating outcome is a conjunction such as a coherent accreted unit with a consistent emplacement age, or an ancient chemical gradient quantitatively consistent with transport under Earth irradiation. A lower gravity misfit alone cannot establish the historical mechanism, because gravity is a nonunique inverse problem.

Choose two hypothetical environments A and B of one chemical element, with line centers ${\mu }_A=284\mathrm{eV}$ and ${\mu }_B=286\mathrm{eV}$. Model each line by a normalized Gaussian function with common standard deviation $\sigma$, and stipulate integrated count weights $w_A=3/4$, $w_B=1/4$. On the binding energy axis $E$, the normalized count density isAll energies in this expression use electronvolts. Each component's integrated area equals its count weight, and $\int I(E)dE=1$. There is no background, statistical counting fluctuation, or additional emission feature in this stipulated model. These choices deliberately isolate photoelectron peak overlap and resolution.

Direct integration, or the law of total expectation applied to the component label, gives the centroidThe centroid lies between the chemical environments and is not either component's binding energy. Expanding $(E-\bar{E}{)}^2$ within each component, the odd cross term integrates to zero. The resulting variance isFor $\sigma =0.25\mathrm{eV}$ the variance is $0.8125{\mathrm{eV}}^2$ and the two maxima are separated. For $\sigma =1.00\mathrm{eV}$ it is $1.7500{\mathrm{eV}}^2$ and there is one maximum. Each Gaussian's full width at half maximum is $2\sqrt{2\ln 2}\sigma$, approximately $0.589$ and $2.355\mathrm{eV}$ respectively. The chemical separation remains $2\mathrm{eV}$ in both cases. The number of visible maxima therefore does not equal the number of chemical components.

One interpretation of this change is additional independent Gaussian instrument blur. The convolution of probability densities adds the variances, so broadening $\sigma =0.25$ to $1$ requires response standard deviation $\sqrt{1-0.25^2}=\sqrt{15}/4\simeq 0.968246\mathrm{eV}$. A normalized response with zero mean preserves the centroid and integrated counts. Real line shapes and backgrounds can depart from this model, and separating hidden components is an inverse problem; a plausible fit does not alone establish uniqueness.

Finally, count weights describe detection. In a simple response model, let $n_i$ denote the number of emitting sites sampled in environment $i$, $s_i$ its effective photoionization cross section, and $T_i$ the probability that its emitted electron contributes to the selected measured line. ThenThe common factor $C$ includes exposure. Here $T_i$ includes escape from the sample, analyzer transmission and detection. Count weights equal site fractions only if the relevant response factors are equal or corrected. They are not automatically molecular mole fractions; different molecules can supply different numbers of the selected emitting sites. In an original example, equal site numbers with $s_A{T}_A=3s_B{T}_B$ give the plotted $3:1$ count ratio despite a $1:1$ site ratio.

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]

Temporal averaging in nail clippings is the mixing of material formed or transported at different times into one measured sample. A clipping has finite length, width and thickness. Even perfect chemical measurement of the whole sample need not resolve the timing of a short exposure.

A primary nail-growth experiment followed file marks near the proximal fold in 22 young adults over one to three months.^[ref] Such displacement measurements constrain the transport clock. Relating a chemical concentration to an earlier exposure requires additional assumptions about incorporation, retention and contamination.

For an original ideal advection model, let material cross a reference at time $s$ with concentration $c(s)$, thereafter retained unchanged. At constant speed $v>0$, a segment between positions $x_1$ and $x_2$ at collection time $t$ contains reference-crossing times from $t-x_2/v$ to $t-x_1/v$. If mass per unit length is uniform, its measured concentration isThis is a shifted moving average, equivalently a convolution with a rectangular window. Its clock is reference crossing; formation in the hidden nail matrix may precede that crossing.

For a 3 mm segment at stipulated speed 0.10 mm/day, $\Delta =30$ days. If a hypothetical incorporated concentration equals $C$ for only five days and zero otherwise, the largest possible clipping average is $5C/30=C/6$. A broad, low measured peak can therefore result from a short, high input without any chemical decay. These values illustrate the model rather than an observed exposure record.

Spatial weighting of nail-clipping exposure records becomes important when thickness, width or density varies: the average is weighted by material mass, not distance alone. Variable growth further distorts the time window. A wider segment generally mixes a longer history, while repeated analyses of the same segment improve analytical precision without shortening that history.

Record clipping extent, material orientation, collection date and how growth was measured. Comparing several separately retained segments can provide more time information than pooling them, but recovering a unique exposure history remains an inverse problem unless the transport and retention model is sufficiently constrained.

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.

Prepare two uniform specimens with the same material, direction and $A_0$, but different deforming lengths $L_1,L_2$. Keep grip engagement, adapters, loading window and measurement configuration equivalent. Determine $q_i=d{x}_i/dF$ separately by linear regression with its own displacement intercept. Subtracting the two series equations eliminates the common machine compliance:This is an inverse problem: one length supplies only one equation for two unknowns, while two distinct lengths separate the length-dependent specimen contribution from the constant contribution. In a graph of compliance against length, the slope is $1/(E{A}_0)$ and the vertical intercept is $C_m$. It is the compliance, not $k_{\mathrm{obs}}$, that is linear in $L$.

The dimensional analysis is revealing: $[q_2-q_1]=m/N$, so $[(L_2-L_1)/(A_0(q_2-q_1))]=N/{\mathrm{m}}^2$. For $L_2>L_1$, the ideal model requires $q_2>q_1$. A negative fitted $C_m$ challenges the common-intercept assumptions or the measurements. Two lengths determine two parameters exactly, so they cannot test the assumed straight line; three or more lengths allow regression residuals to reveal departures. Nearly equal lengths amplify slope errors through the small difference $q_2-q_1$. If the actual apparatus compliances differ by $\Delta C_m$, the inferred length slope becomes $1/(E{A}_0)+\Delta C_m/(L_2-L_1)$.

Choose two point-like astronomical sources with $({\theta }_1,I_1)=(-0.00625,1)$ and $({\theta }_2,I_2)=(0.009375,0.65)$, where the angles are in radians and strengths use arbitrary common units. Their separation is $\Delta \theta =0.015625$ radians. These are invented model parameters, not historical measurements. Their complex visibility follows immediately from the preceding derivation, and its squared magnitude isAt $u=0$, $V=1.65$; at $u=400$, $V=-1+0.65i$; at $u=800$, $V=0.35$. The unequal strengths prevent the complex visibility magnitude from reaching zero. Its modulation contains the source separation, while a common displacement of both sources changes wave phase without changing this magnitude. Amplitude alone loses that location information.

The sparse case samples $u=\pm 12.5k$ for $k\in \{2,5,9,14,20,28,36,44,52,60,64\}$; the denser case samples every $k=1,\dots ,64$, also with both signs. All weights are one. Both reach $|u|=800$, so their shortest fringe period is the same $0.00125$ radians. Their dirty beams and false subsidiary features differ because the intermediate spatial frequencies differ. Neither case includes $u=0$ or frequencies above that maximum. The figure uses a periodic angular interval of length $0.08$ radians; this artificial boundary makes a discrete Fourier transform calculation exact for the chosen grid, rather than modelling an all-sky observation.

The plotted dirty image is the sum of two shifted dirty beams, not a fitted or deconvolved estimate of the source. Adding samples in this example reduces prominent spurious peaks; it does not establish a universal monotonic improvement for every weighting or sky. A missing zero baseline leaves the periodic dirty image with zero mean, so it cannot by itself recover the total positive brightness. Missing short spacings also weaken constraints on extended smooth emission.

Reconstruction is an inverse problem. A perturbation whose Fourier transform vanishes at every sampled coordinate can be added without changing the measurements: incomplete coverage leaves a null space. Nonnegative brightness, a restricted source region, or a model with a small number of components can reduce ambiguity, but each is additional information. Removing dirty beam artefacts means estimating a sky consistent with both data and such assumptions; a narrow central peak alone does not certify a unique image. Measurement error, uncertain calibration and source variability add further limitations.

In variable projection, a prediction has the form $\mathbf{G}(\mathbf{q})\mathbf{m}$, nonlinear in $\mathbf{q}$ but linear in $\mathbf{m}$. For each trial $\mathbf{q}$, solve the inner weighted least squares problem for $\mathbf{m}$, then optimize the resulting residual over $\mathbf{q}$. This reduces the nonlinear dimension without pretending that the whole inverse problem is linear.

Will J. Merry coauthored notes with Gabriel Paternain on inverse problems in geometry and dynamical systems. They develop geodesic flows, the Anosov flows and Livšic theorem as tools for reconstruction. A recurring idea is to infer information about a function or a Riemannian metric from integrals along trajectories, rather than from direct measurements at every interior point.^[ref]

X-ray astronomy infers the properties of celestial sources from their X-rays. It requires instruments above the strongly absorbing layers of Earth's terrestrial atmosphere. Riccardo Giacconi's 2002 Nobel lecture connects the early rocket detections to later grazing-incidence X-ray optics, which collect photons over a large entrance area onto a small detector.^[ref]

The measurement is an inverse problem: a count rate mixes incident photon arrival rate, detector response, and background. A point-like astronomical source produces a direction-dependent excess as the instrument scans; diffuse X-ray background contributes over many directions. Neither a count excess nor a narrow feature by itself determines the emission mechanism. The useful progression is from detecting an excess, to locating it, to measuring its spectrum and light curve.

X-ray crystallographic structure inference is an inverse problem: infer a crystal structure from measured X-ray diffraction. A proposed unit cell and atomic basis of a crystal predict a pattern; structure determination asks which proposals survive comparison with that pattern. The useful separation is between directions and strengths. Peak positions constrain the reciprocal lattice and hence interplanar spacings; relative wave intensities constrain the motif through the crystal basis structure factor. Knowing a repeat distance does not locate all the chemical atoms within that repeat.

The lattice constant from two indexed reflections example isolates the geometric inference. Basis intensities and phase ambiguity then holds the translations fixed while changing the motif. Even complete, noiseless wave intensity data need not determine all wave phases: this is the phase problem. Chemical composition, symmetry, and additional measurements can supply constraints beyond the original pattern.^[ref]

For a fixed homogeneous absorber in a narrow-beam geometry, define the X-ray spectral attenuation operator on L1 space over a bounded positive energy interval. A physical input $f(E)\ge 0$ is an integrable function giving incident photon rate density. Let its linear attenuation coefficient $\mu (E)\ge 0$ be a measurable function, with $x\ge 0$. The transmitted uncollided density isHere $w(E)\ge 0$ is a fixed bounded detector response described by a measurable function; background has been subtracted. This is a positive linear operator on L1 space. Since $0\le e^{-\mu (E)x}\le 1$, it cannot increase the integrated absolute incident rate. Multiplication of exponentials proves $A_{x+y}=A_x{A}_y$ for successive layers of the same material. Ignore secondary radiation entering the detector, saturation and changes to the source. Discrete spectral lines use the finite-sum version below instead of an ordinary rate density.

For a chosen filter, stipulate ${\mu }_{\alpha }=0.20\mathrm{m}{\mathrm{m}}^{-1}$, ${\mu }_{\beta }=0.80\mathrm{m}{\mathrm{m}}^{-1}$ and input line rates $(N_{\alpha },N_{\beta })=(4,1)$ in a common arbitrary rate unit. Assume equal calibrated detector weight for both lines. At $x=2.0\mathrm{mm}$,The selected $8.5\mathrm{keV}$ line's fraction rises from $4/5=0.800$ to $0.929974$, but its absolute rate falls by about $33.0\%$. Filtering improves purity at the cost of signal. It multiplies existing components and cannot create a new emitted spectral line.

With just one integrated reading, $S=N_{\alpha }+N_{\beta }$, the two unknown line rates are underdetermined. If the source is unchanged between an empty-path reading and a filtered reading, the ideal inverse problem becomesThe determinant $e^{-1.6}-e^{-0.4}\simeq -0.468424$ is nonzero, so these two ideal readings determine the two stipulated line rates uniquely. Identical coefficients would give zero determinant and destroy that separation. This small linear algebra example explains why absorber contrast supplies information, while also showing its limit: unknown scattered components or unknown detector weights add unknowns that two readings cannot determine.