In the near future, ultra deep observations of galaxy clusters with HST or JWST will uncover $300-1000$ lensed multiple images, increasing the current count per cluster by up to an order of magnitude. This will further refine our view of clusters, leading to a more accurate and precise mapping of the total and dark matter distribution in clusters, and enabling a better understanding of background galaxy population and their luminosity functions. However, to effectively use that many images as input to lens inversion will require a re-evaluation of, and possibly upgrades to the existing methods. In this paper we scrutinize the performance of the free-form lens inversion method Grale in the regime of $150-1000$ input images, using synthetic massive galaxy clusters. Our results show that with an increasing number of input images, Grale produces improved reconstructed mass distributions, with the fraction of the lens plane recovered at better than $10\%$ accuracy increasing from $40-50\%$ for $\sim\!\!150$ images to $65\%$ for $\sim\!1000$ images. The reconstructed time delays imply a more precise measurement of $H_0$, with $\lesssim 1\%$ bias. While the fidelity of the reconstruction improves with the increasing number of multiple images used as model constraints, $\sim 150$ to $\sim 1000$, the lens plane rms deteriorates from $\sim 0.11''$ to $\sim 0.28''$. Since lens plane rms is not necessarily the best indicator of the quality of the mass reconstructions, looking for an alternative indicator is warranted.

The information about the mass density of galaxy clusters provided by the gravitational lens effect has inspired many inversion techniques. In this article, updates to the previously introduced method in Grale are described, and explored in a number of examples. The first looks into a different way of incorporating time delay information, not requiring the unknown source position. It is found that this avoids a possible bias that leads to "over-focusing" the images, i.e. providing source position estimates that lie in a considerably smaller region than the true positions. The second is inspired by previous reconstructions of the cluster of galaxies MACS J1149.6+2223, where a multiply-imaged background galaxy contained a supernova, SN Refsdal, of which four additional images were produced by the presence of a smaller cluster galaxy. The inversion for the cluster as a whole, was not able to recover sufficient detail interior to this quad. We show how constraints on such different scales, from the entire cluster to a single member galaxy, can now be used, allowing such small scale substructures to be resolved. Finally, the addition of weak lensing information to this method is investigated. While this clearly helps recover the environment around the strong lensing region, the mass sheet degeneracy may make a full strong and weak inversion difficult, depending on the quality of the ellipticity information at hand. We encounter ring-like structure at the boundary of the two regimes, argued to be the result of combining strong and weak lensing constraints, possibly affected by degeneracies.

Abell 2744, a massive Hubble Frontier Fields merging galaxy cluster with many multiple images in the core has been the subject of many lens inversions using different methods. While most existing studies compare various inversion methods, we focus on a comparison of reconstructions that use different input lensing data. Since the quantity and quality of lensing data is constantly improving, it makes sense to ask if the estimated uncertainties are robust against changes in the data. We address this question using free-form Grale, which takes only image information as input, and nothing pertaining to cluster galaxies. We reconstruct Abell 2744 using two sets of strong lensing data from the Hubble Frontier Fields community. Our first and second reconstructions use 55 and 91 images, respectively, and only 10 of the 91 images have the same positions and redshifts as in the first reconstruction. Comparison of the two mass maps shows that Grale uncertainties are robust against these changes, as well as small modifications in the inversion routine. Additionally, applying the methods used in Sebesta et al. (2016) for MACS J0416, we conclude that, in a statistical sense, light follows mass in Abell 2744, with brighter galaxies clustering stronger with the recovered mass than the fainter ones. We also show that the faintest galaxies are anti-correlated with mass, which is likely the result of light contamination from bright galaxies, and lensing magnification bias acting on galaxies background to the cluster.

We investigate to which precision local magnification ratios, $\mathcal{J}$, ratios of convergences, $f$, and reduced shears, $g = (g_{1}, g_{2})$, can be determined model-independently for the five resolved multiple images of the source at $z_\mathrm{s}=1.675$ in CL0024. We also determine if a comparison to the respective results obtained by the parametric modelling program Lenstool and by the non-parametric modelling program Grale can detect biases in the lens models. For these model-based approaches we additionally analyse the influence of the number and location of the constraints from multiple images on the local lens properties determined at the positions of the five multiple images of the source at $z_\mathrm{s}=1.675$. All approaches show high agreement on the local values of $\mathcal{J}$, $f$, and $g$. We find that Lenstool obtains the tightest confidence bounds even for convergences around one using constraints from six multiple image systems, while the best Grale model is generated only using constraints from all multiple images with resolved brightness features and adding limited small-scale mass corrections. Yet, confidence bounds as large as the values themselves can occur for convergences close to one in all approaches. Our results are in agreement with previous findings, supporting the light-traces-mass assumption and the merger hypothesis for CL0024. Comparing the three different approaches allows to detect modelling biases. Given that the lens properties remain approximately constant over the extension of the image areas covered by the resolvable brightness features, the model-independent approach determines the local lens properties to a comparable precision but within less than a second. (shortened)

With upcoming (continuum) surveys of high-resolution radio telescopes, detection rates of fast radio bursts (FRBs) might approach $10^5$ per sky per day by future extremely large observatories, such as the possible extension of the Square Kilometer Array (SKA) to a phase 2 array. Depending on the redshift distribution of FRBs and using the repeating FRB121102 as a model, we calculate a detection rate of multiply-imaged FRBs with their multiply-imaged hosts caused by the distribution of galaxy-cluster scale gravitational lenses of the order of $10^{-4}$ per square degree per year for a minimum total flux of the host of 10 $\mu$Jy at 1.4 GHz for SKA phase 2. Our comparison of estimated detection rates for quasars, supernovae, gamma ray bursts, and FRBs shows that multiple images of FRBs could be more numerous than those of gamma ray bursts and supernovae and as numerous as multiple images of quasars. Time delays between the multiple images of an FRB break degeneracies in model-based and model-independent lens reconstructions as other time-varying sources do, yet without a microlensing bias as FRBs are more point-like and have shorter duration times. We estimate the relative imprecision of FRB time-delay measurements to be $10^{-10}$ for time delays on the order of 100 days for galaxy-cluster scale lenses, yielding more precise (local) lens properties than time delays from the other time-varying sources. Using the lens modelling software Grale, we show the increase in accuracy and precision of the reconstructed scaled surface mass density map of a simulated cluster-scale lens when adding time delays for one set of multiple images to the set of observational constraints.

Multiple image gravitational lensing systems with measured time delays provide a promising one-step method for determining $H_0$. MACS J1149, which lenses SN Refsdal into a quad S1-S4, and two other widely separated images, SX and SY, is a perfect candidate. If time delays are pinned down, the remaining uncertainty arises from the mass distribution in the lens. In MACS J1149, the mass in the relevant lens plane region can be constrained by (i) many multiple images, (ii) the mass of the galaxy splitting S1-S4 (which, we show, is correlated with $H_0$), (iii) magnification of SX (also correlated with $H_0$), and (iv) prior assumptions on the mass distribution. Our goal is not to estimate $H_0$, but to understand its error budget, i.e., estimate uncertainties associated with each of these constraints. Using multiple image positions alone, yields very large uncertainty, despite the fact that the position of SX is recovered to within $\!\le\!0.036$" (rms $\!\le\!0.36$") by GRALE lens inversion. Fixing the mass of the galaxy that splits S1-S4 reduces $1\sigma$ uncertainties to $\sim 23\%$, while fixing the magnification of SX yields $1\sigma$ uncertainties of $32\%$. We conclude that smaller uncertainties, of order few percent, are a consequence of imposing prior assumptions on the shapes of the galaxy and cluster mass distributions, which may or may not apply in a highly non-equilibrium environment of a merging cluster. We propose that if a measurement of $H_0$ is to be considered reliable, it must be supported by a wide range of lens inversion methods.

MACS J0717 is the most massive and extended of the Hubble Frontier Field clusters. It is one of the more difficult clusters to model, and we argue that this is in part due to the line of sight structure (LoS) at redshifts beyond 2. We show that the Grale mass reconstruction based on sources at 3<z_s<4.1 has at least 10^{13}M_sun more mass than that based on nearby sources, z_s<2.6, and attribute the excess mass to a putative LoS, which is at least 75" from the cluster center. Furthermore, the lens-model fitted z_s's of the recent Kawamata et al. reconstruction are biased systematically low compared to photometric z_s's, and the bias is a function of images' distance from the cluster center. We argue that these mimic the effect of LoS. We conclude that even in the presence of 100-200 images, lens-model adjusted source redshifts can conceal the presence of LoS, demonstrating the existence of degeneracies between z_s and (sub)structure. Also, a very good fit to image positions is not a sufficient condition for having a high fidelity mass map: Kawamata et al. obtain an rms of 0.52" for 173 images of 60 sources; our Grale reconstruction of the exact same data yields a somewhat different map, but similarly low rms, 0.62". In contrast, a Grale model that uses reasonable, but fixed z_s gives a worse rms of 1.28" for 44 sources with 126 images. Unaccounted for LoS can bias the mass map, affecting the magnification and luminosity function estimates of high redshift sources.

We present integral field spectroscopy of galaxy cluster Abell 3827, using ALMA and VLT/MUSE. It reveals an unusual configuration of strong gravitational lensing in the cluster core, with at least seven lensed images of a single background spiral galaxy. Lens modelling based on HST imaging had suggested that the dark matter associated with one of the cluster's central galaxies may be offset. The new spectroscopic data enable better subtraction of foreground light, and better identification of multiple background images. The inferred distribution of dark matter is consistent with being centered on the galaxies, as expected by LCDM. Each galaxy's dark matter also appears to be symmetric. Whilst we do not find an offset between mass and light (suggestive of self-interacting dark matter) as previously reported, the numerical simulations that have been performed to calibrate Abell 3827 indicate that offsets and asymmetry are still worth looking for in collisions with particular geometries. Meanwhile, ALMA proves exceptionally useful for strong lens image identifications.

Using the power of gravitational lensing magnification by massive galaxy clusters, the Hubble Frontier Fields provide deep views of six patches of the high redshift Universe. The combination of deep Hubble imaging and exceptional lensing strength has revealed the greatest numbers of multiply-imaged galaxies available to constrain models of cluster mass distributions. However, even with $\mathcal{O}(100)$ images per cluster, the uncertainties associated with the reconstructions are not negligible. The goal of this paper is to show the diversity of model magnification predictions. We examine 7 and 9 mass models of Abell 2744 and MACS J0416, respectively, submitted to the Mikulski Archive for Space Telescopes for public distribution in September 2015. The dispersion between model predictions increases from 30% at common low magnifications ($\mu\sim2$) to 70% at rare high magnifications ($\mu\sim40$). MACS J0416 exhibits smaller dispersions than Abell 2744 for $2<\mu<10$. We show that magnification maps based on different lens inversion techniques typically differ from each other by more than their quoted statistical errors. This suggests that some models underestimate the true uncertainties, which are primarily due to various lensing degeneracies. Though the exact mass sheet degeneracy is broken, its generalized counterpart is not broken at least in Abell 2744. Other, local degeneracies are also present in both clusters. Our comparison of models is complementary to the comparison of reconstructions of known synthetic mass distributions. By focusing on observed clusters, we can identify those that are best constrained, and therefore provide the clearest view of the distant Universe.

We reconstruct the projected mass distribution of a massive merging Hubble Frontier Fields cluster MACSJ0416 using the genetic algorithm based free-form technique called Grale. The reconstructions are constrained by 149 lensed images identified by Jauzac et al. using HFF data. No information about cluster galaxies or light is used, which makes our reconstruction unique in this regard. Using visual inspection of the maps, as well as galaxy-mass correlation functions we conclude that overall light does follow mass. Furthermore, the fact that brighter galaxies are more strongly clustered with mass is an important confirmation of the standard biasing scenario in galaxy clusters. On the smallest scales, approximately less than a few arcseconds, the resolution afforded by 149 images is still not sufficient to confirm or rule out galaxy-mass offsets of the kind observed in ACO 3827. We also compare the mass maps of MACSJ0416 obtained by three different groups: Grale, and two parametric Lenstool reconstructions from the CATS and Sharon/Johnson teams. Overall, the three agree well; one interesting discrepancy between Grale and Lenstool galaxy-mass correlation functions occurs on scales of tens of kpc and may suggest that cluster galaxies are more biased tracers of mass than parametric methods generally assume.

Gravitational lensing by clusters of galaxies offers a powerful probe of their structure and mass distribution. Deriving a lens magnification map for a galaxy cluster is a classic inversion problem and many methods have been developed over the past two decades to solve it. Several research groups have developed techniques independently to map the predominantly dark matter distribution in cluster lenses. While these methods have all provided remarkably high precision mass maps, particularly with exquisite imaging data from the Hubble Space Telescope (HST), the reconstructions themselves have never been directly compared. In this paper, we report the results of comparing various independent lens modeling techniques employed by individual research groups in the community. Here we present for the first time a detailed and robust comparison of methodologies for fidelity, accuracy and precision. For this collaborative exercise, the lens modeling community was provided simulated cluster images -- of two clusters Ares and Hera -- that mimic the depth and resolution of the ongoing HST Frontier Fields. The results of the submitted reconstructions with the un-blinded true mass profile of these two clusters are presented here. Parametric, free-form and hybrid techniques have been deployed by the participating groups and we detail the strengths and trade-offs in accuracy and systematics that arise for each methodology. We note in conclusion that lensing reconstruction methods produce reliable mass distributions that enable the use of clusters as extremely valuable astrophysical laboratories and cosmological probes.

The Hubble Frontier Fields (HFF) are six clusters of galaxies, all showing indications of recent mergers, which have recently been observed for lensed images. As such they are the natural laboratories to study the merging history of galaxy clusters. In this work, we explore the 2D power spectrum of the mass distribution $P_{\rm M}(k)$ as a measure of substructure. We compare $P_{\rm M}(k)$ of these clusters (obtained using strong gravitational lensing) to that of $\Lambda$CDM simulated clusters of similar mass. To compute lensing $P_{\rm M}(k)$, we produced free-form lensing mass reconstructions of HFF clusters, without any light traces mass (LTM) assumption. The inferred power at small scales tends to be larger if (i)~the cluster is at lower redshift, and/or (ii)~there are deeper observations and hence more lensed images. In contrast, lens reconstructions assuming LTM show higher power at small scales even with fewer lensed images; it appears the small scale power in the LTM reconstructions is dominated by light information, rather than the lensing data. The average lensing derived $P_{\rm M}(k)$ shows lower power at small scales as compared to that of simulated clusters at redshift zero, both dark-matter only and hydrodynamical. The possible reasons are: (i)~the available strong lensing data are limited in their effective spatial resolution on the mass distribution, (ii)~HFF clusters have yet to build the small scale power they would have at $z\sim 0$, or (iii)~simulations are somehow overestimating the small scale power.

Galaxy cluster Abell 3827 hosts the stellar remnants of four almost equally bright elliptical galaxies within a core of radius 10kpc. Such corrugation of the stellar distribution is very rare, and suggests recent formation by several simultaneous mergers. We map the distribution of associated dark matter, using new Hubble Space Telescope imaging and VLT/MUSE integral field spectroscopy of a gravitationally lensed system threaded through the cluster core. We find that each of the central galaxies retains a dark matter halo, but that (at least) one of these is spatially offset from its stars. The best-constrained offset is 1.62+/-0.48kpc, where the 68% confidence limit includes both statistical error and systematic biases in mass modelling. Such offsets are not seen in field galaxies, but are predicted during the long infall to a cluster, if dark matter self-interactions generate an extra drag force. With such a small physical separation, it is difficult to definitively rule out astrophysical effects operating exclusively in dense cluster core environments - but if interpreted solely as evidence for self-interacting dark matter, this offset implies a cross-section sigma/m=(1.7+/-0.7)x10^{-4}cm^2/g x (t/10^9yrs)^{-2}, where t is the infall duration.

Gravitational lensing time delays are well known to depend on cosmological parameters, but they also depend on the details of the mass distribution of the lens. It is usual to model the mass distribution and use time-delay observations to infer cosmological parameters, but it is naturally also possible to take the cosmological parameters as given and use time delays as constraints on the mass distribution. This paper develops a method to isolate what exactly those constraints are, using a principal-components analysis of ensembles of free-form mass models. We find that time delays provide tighter constraints on the distribution of matter in the very high dense regions of the lensing clusters. We apply it to the cluster lens SDSS J1004+4112, whose rich lensing data includes two time delays. We find, assuming a concordance cosmology, that the time delays constrain the central region of the cluster to be rounder and less lopsided than would be allowed by lensed images alone. This detailed information about the distribution of the matter is very useful for studying the dense regions of the galaxy clusters which are very difficult to study with direct measurements. A further time-delay measurement, which is expected, will make this system even more interesting.

We have made mass maps of three strong-lensing clusters, Abell 3827, Abell 2218 and Abell 1689, in order to test for mass-light offsets. The technique used is GRALE, which enables lens reconstruction with minimal assumptions, and specifically with no information about the cluster light being given. In the first two of these clusters, we find local mass peaks in the central regions that are displaced from the nearby galaxies by a few to several kpc. These offsets {\em could\/} be due to line of sight structure unrelated to the clusters, but that is very unlikely, given the typical levels of chance line-of-sight coincidences in $\Lambda CDM$ simulations --- for Abell 3827 and Abell 2218 the offsets appear to be intrinsic. In the case of Abell 1689, we see no significant offsets in the central region, but we do detect a possible line of sight structure: it appears only when sources at $z\ga 3$ are used for reconstructing the mass. We discuss possible origins of the mass-galaxy offsets in Abell 3827 and Abell 2218: these include pure gravitational effects like dynamical friction, but also non-standard mechanisms like self-interacting dark-matter.

The inversion of gravitational lens systems is hindered by the fact that multiple mass distributions are often equally compatible with the observed properties of the images. Besides using clear examples to illustrate the effect of the so-called monopole and mass sheet degeneracies, this article introduces the most general form of said mass sheet degeneracy. While the well known version of this degeneracy rescales a single source plane, this generalization allows any number of sources to be rescaled. Furthermore, it shows how it is possible to rescale each of those sources with a different scale factor. Apart from illustrating that the mass sheet degeneracy is not broken by the presence of multiple sources at different redshifts, it will become apparent that the newly constructed mass distribution necessarily alters the existing mass density precisely at the locations of the images in the lens system, and that this change in mass density is linked to the factors with which the sources were rescaled. Combined with the fact that the monopole degeneracy introduces a large amount of uncertainty about the density in between the images, this means that both degeneracies are in fact closely related to substructure in the mass distribution. An example simulated lensing situation based on an elliptical version of a Navarro-Frenk-White profile explicitly shows that such degeneracies are not easily broken by observational constraints, even when multiple sources are present. Instead, the fact that each lens inversion method makes certain assumptions, implicit or explicit, about the smoothness of the mass distribution means that in practice the degeneracies are broken in an artificial manner rather than by observed properties of the lens system.

We present a strong-lensing analysis of the galaxy cluster MS 1358.4+6245 ($z=0.33$), in deep 6-band ACS/HST imaging. In addition to the well-studied system at $z=4.92$, our modelling method uncovers 19 new multiply-lensed images so that a total of 23 images and their redshifts are used to accurately constrain the inner mass distribution. We derive a relatively shallow inner mass profile, $d\log \Sigma/d\log r\simeq -0.33 \pm0.05$ ($r<200$ kpc), with a much higher magnification than estimated previously by models constrained only by the $z=4.92$ system. Using these many new images we can apply a non-parametric adaptive-grid method, which also yields a shallow mass profile without prior assumptions, strengthening our conclusions. The total magnification of the $z_s=4.92$ galaxy is high, about a $\sim100\times$ over its four images, so that the inferred source size, luminosity and star-formation rate are about $\sim5\times$ smaller than previous estimates, corresponding to a dwarf-sized galaxy of radius $\simeq1$ kpc. A detailed image of the interior morphology of the source is generated with a high effective resolution of only $\simeq$50 pc, thanks to the high magnification and to the declining angular diameter distance above $z\sim1.5$ for the standard cosmology, so that this image apparently represents the best resolved object known at high redshift.

The inner mass-profile of the relaxed cluster Abell 1703 is analysed by two very different strong-lensing techniques applied to deep ACS and WFC3 imaging. Our parametric method has the accuracy required to reproduce the many sets of multiple images, based on the assumption that mass approximately traces light. We test this assumption with a fully non-parametric, adaptive grid method, with no knowledge of the galaxy distribution. Differences between the methods are seen on fine scales due to member galaxies which must be included in models designed to search for lensed images, but on the larger scale the general distribution of dark matter is in good agreement, with very similar radial mass profiles. We add undiluted weak-lensing measurements from deep multi-colour Subaru imaging to obtain a fully model-independent mass profile out to the virial radius and beyond. Consistency is found in the region of overlap between the weak and strong lensing, and the full mass profile is well-described by an NFW model of a concentration parameter, $c_{\rm vir}\simeq 7.15\pm0.5$ (and $M_{vir}\simeq 1.22\pm0.15 \times 10^{15}M_{\odot}/h$). Abell 1703 lies above the standard $c$--$M$ relation predicted for the standard $\Lambda$CDM model, similar to other massive relaxed clusters with accurately determined lensing-based profiles.

In this article we study the well-known strong lensing system SDSS J1004+4112. Not only does it host a large-separation lensed quasar with measured time-delay information, but several other lensed galaxies have been identified as well. A previously developed strong lens inversion procedure that is designed to handle a wide variety of constraints, is applied to this lensing system and compared to results reported in other works. Without the inclusion of a tentative central image of one of the galaxies as a constraint, we find that the model recovered by the other constraints indeed predicts an image at that location. An inversion which includes the central image provides tighter constraints on the shape of the central part of the mass map. The resulting model also predicts a central image of a second galaxy where indeed an object is visible in the available ACS images. We find masses of 2.5x10^13 M_O and 6.1x10^13 M_O within a radius of 60 kpc and 110 kpc respectively, confirming the results from other authors. The resulting mass map is compatible with an elliptical generalization of a projected NFW profile, with r_s = 58_{-13}^{+21} arcsec and c_vir = 3.91 +/- 0.74. The orientation of the elliptical NFW profile follows closely the orientation of the central cluster galaxy and the overall distribution of cluster members.

The cluster lens Cl 0024+1654 is undoubtedly one of the most beautiful examples of strong gravitational lensing, providing five large images of a single source with well-resolved substructure. Using the information contained in the positions and the shapes of the images, combined with the null space information, a non-parametric technique is used to infer the strong lensing mass map of the central region of this cluster. This yields a strong lensing mass of 1.60x10^14 M_O within a 0.5' radius around the cluster center. This mass distribution is then used as a case study of the monopole degeneracy, which may be one of the most important degeneracies in gravitational lensing studies and which is extremely hard to break. We illustrate the monopole degeneracy by adding circularly symmetric density distributions with zero total mass to the original mass map of Cl 0024+1654. These redistribute mass in certain areas of the mass map without affecting the observed images in any way. We show that the monopole degeneracy and the mass-sheet degeneracy together lie at the heart of the discrepancies between different gravitational lens reconstructions that can be found in the literature for a given object, and that many images/sources, with an overall high image density in the lens plane, are required to construct an accurate, high-resolution mass map based on strong-lensing data.

The inversion of a gravitational lens system is, as is well known, plagued by the so-called mass-sheet degeneracy: one can always rescale the density distribution of the lens and add a constant-density mass-sheet such that the, also properly rescaled, source plane is projected onto the same observed images. For strong lensing systems, it is often claimed that this degeneracy is broken as soon as two or more sources at different redshifts are available. This is definitely true in the strict sense that it is then impossible to add a constant-density mass-sheet to the rescaled density of the lens without affecting the resulting images. However, often one can easily construct a more general mass distribution -- instead of a constant-density sheet of mass -- which gives rise to the same effect: a uniform scaling of the sources involved without affecting the observed images. We show that this can be achieved by adding one or more circularly symmetric mass distributions, each with its own center of symmetry, to the rescaled mass distribution of the original lens. As it uses circularly symmetric distributions, this procedure can lead to the introduction of ring shaped features in the mass distribution of the lens. In this paper, we show explicitly how degenerate inversions for a given strong lensing system can be constructed. It then becomes clear that many constraints are needed to effectively break this degeneracy.

Galaxies acting as gravitational lenses are surrounded by, at most, a handful of images. This apparent paucity of information forces one to make the best possible use of what information is available to invert the lens system. In this paper, we explore the use of a genetic algorithm to invert in a non-parametric way strong lensing systems containing only a small number of images. Perhaps the most important conclusion of this paper is that it is possible to infer the mass distribution of such gravitational lens systems using a non-parametric technique. We show that including information about the null space (i.e. the region where no images are found) is prerequisite to avoid the prediction of a large number of spurious images, and to reliably reconstruct the lens mass density. While the total mass of the lens is usually constrained within a few percent, the fidelity of the reconstruction of the lens mass distribution depends on the number and position of the images. The technique employed to include null space information can be extended in a straightforward way to add additional constraints, such as weak lensing data or time delay information.

We present a non-parametric technique to infer the projected-mass distribution of a gravitational lens system with multiple strong-lensed images. The technique involves a dynamic grid in the lens plane on which the mass distribution of the lens is approximated by a sum of basis functions, one per grid cell. We used the projected mass densities of Plummer spheres as basis functions. A genetic algorithm then determines the mass distribution of the lens by forcing images of a single source, projected back onto the source plane, to coincide as well as possible. Averaging several tens of solutions removes the random fluctuations that are introduced by the reproduction process of genomes in the genetic algorithm and highlights those features common to all solutions. Given the positions of the images and the redshifts of the sources and the lens, we show that the mass of a gravitational lens can be retrieved with an accuracy of a few percent and that, if the sources sufficiently cover the caustics, the mass distribution of the gravitational lens can also be reliably retrieved. A major advantage of the algorithm is that it makes full use of the information contained in the radial images, unlike methods that minimise the residuals of the lens equation, and is thus able to accurately reconstruct also the inner parts of the lens.