N+2 Mapmaking in Polarized CMB

• N+2 mapmaking is an augmented approach that estimates N detector-specific temperature maps alongside two common Q and U polarization fields.
• It employs generalized least-squares estimation to mitigate temperature-to-polarization leakage caused by bandpass mismatches.
• Effective implementation relies on proper scan strategies and conditioning, with alternatives like horn maps offering practical noise-leakage compromises.

Searching arXiv for relevant papers on “N+2 Mapmaking” and closely related mapmaking formulations. First, I’ll search for the exact phrase and then for closely related polarized/CMB and reduced-mode mapmaking work. N+2 Mapmaking is an augmented mapmaking formulation in which the unknown state is enlarged beyond a single coadded map. In the explicit polarized CMB sense introduced under that name, one solves for $N$ detector-specific temperature maps together with two shared linear-polarization Stokes parameters, $Q$ and $U$, so that each pixel carries $(I_1,\dots,I_N,Q,U)$ rather than the standard $(I,Q,U)$. The immediate motivation is bandpass mismatch: detectors nominally assigned to one frequency channel do not in general see the same effective temperature sky, and forcing them into a common intensity map can project detector-dependent intensity structure into polarization. In the surrounding literature, the same phrase is also used more loosely for related mapmaking strategies that append nuisance variables, low-order modes, or latent fields to a base inverse problem rather than solving only for a single map (Galloway et al., 26 Sep 2025).

1. Definition and scope

In its strictest usage, N+2 mapmaking denotes a multi-detector polarized mapmaker that solves for one temperature-like map per detector and two common polarization fields. The name is literal: $N$ is the number of detector-specific temperature maps, and $+2$ refers to the shared $Q$ and $U$ maps. The method is presented as producing a statistically coherent set of physically meaningful per-detector temperature maps, each with slightly different bandpasses as defined by each detector, together with coadded polarization maps (Galloway et al., 26 Sep 2025).

The term also acquired a broader methodological meaning in adjacent work. In CMB and related inverse problems, it is naturally associated with augmented systems in which the map is solved jointly with offsets, gains, baselines, messenger fields, or low-order harmonic modes rather than as a standalone sky estimate. In that broader sense, Planck destriping with baselines, calibration-coupled mapmaking with ring gains and offsets, messenger-field solvers, and reduced-mode harmonic mapmakers are all direct precursors or analogues, even when the phrase “N+2 mapmaking” is not used explicitly (Collaboration et al., 2015).

A narrower but important ambiguity is that “mapmaking” itself is field-dependent. In observational cosmology it denotes a linear inverse problem from timestream or visibility data to sky maps; in geometric topology and combinatorics it can denote the construction or enumeration of maps on surfaces, or the dimension-efficient construction of $k$-regular maps. Some of those literatures also contain explicit “$Q$0” shifts, but the meaning of the shift is then different from the polarized-CMB usage (Buczyński et al., 2015).

2. Polarized CMB formulation

The detector-level signal model introduced for N+2 mapmaking is

$Q$1

where detector $Q$2 contributes to its own temperature map $Q$3 while all detectors share the same $Q$4 and $Q$5. The per-pixel unknown vector is therefore

$Q$6

and the generalized pointing matrix contains one detector-selector column for each $Q$7, followed by the usual polarization columns $Q$8 and $Q$9. With diagonal detector-noise covariance,

$U$0

the map estimate is the standard generalized least-squares solution

$U$1

The innovation lies not in the estimator class but in the state vector and hence in the structure of $U$2 (Galloway et al., 26 Sep 2025).

This parameterization is closely related to the “spurious mapmaking” algorithm pioneered by the WMAP team, but the basis is different. Spurious-map approaches solve for a common intensity map plus one or more mismatch fields, whereas N+2 mapmaking solves directly for all detector-specific temperatures. The two descriptions span comparable temperature subspaces, but N+2 makes the temperature degrees of freedom physically interpretable as per-detector bandpass-weighted skies (Galloway et al., 26 Sep 2025).

The physical motivation is temperature-to-polarization leakage suppression. If detector bandpasses differ, then bright foregrounds or line emission produce genuine detector-dependent temperature signals. In a conventional $U$3 solve, those differences have no dedicated intensity subspace and can project into $U$4. In N+2 mapmaking they are absorbed by the $U$5. The paper demonstrates this with a simulation in which a $U$6 mK detector-specific offset injected into one detector produces a visible leakage pattern in standard coadded mapmaking but is suppressed by more than six orders of magnitude in the N+2 formulation (Galloway et al., 26 Sep 2025).

3. Conditioning, scan strategy, and empirical behavior

The main practical constraint is conditioning. The relevant matrix is the per-pixel coupling matrix $U$7, whose off-diagonal temperature-polarization blocks are proportional to weighted sums of $U$8 and $U$9. With nearly uniform polarization-angle coverage, those averages approach zero, the matrix becomes close to block diagonal, and detector-specific temperatures decouple cleanly from the shared $(I_1,\dots,I_N,Q,U)$0. With poor angle coverage, near-degenerate combinations of detector-specific intensity and shared polarization appear, and the inversion becomes noisy (Galloway et al., 26 Sep 2025).

This issue is decisive for Planck LFI 30 GHz. Applied to detectors 27M, 27S, 28M, and 28S within the Commander3 framework, the full detector-level solve is reported to be too poorly cross-linked to allow a clean separation between temperature and polarization. The mean condition number of the per-detector coupling matrix is about $(I_1,\dots,I_N,Q,U)$1, compared with about $(I_1,\dots,I_N,Q,U)$2 for the coadded case, which the paper interprets as effectively boosting the white-noise variance of the individual-detector solve by about a factor of $(I_1,\dots,I_N,Q,U)$3. The resulting $(I_1,\dots,I_N,Q,U)$4 maps are correspondingly degraded (Galloway et al., 26 Sep 2025).

The same Planck analysis suggests a compromise basis. Detectors within a single horn are strongly anti-correlated, and their temperature-polarization coupling patterns have nearly identical morphology with opposite sign. The paper therefore argues that solving for horn maps, rather than individual detector maps, may provide an optimal compromise between noise and temperature-to-polarization leakage minimization. This preserves some bandpass granularity without paying the full conditioning penalty of one intensity map per detector (Galloway et al., 26 Sep 2025).

By contrast, the method behaves as intended in a simulated scan with dramatically improved angle coverage. Replacing the detector polarization angle at each sample with a random orientation, as a proxy for an ideal fast rotating half-wave plate, makes the problematic coupling terms average toward zero. In that setting the N+2 solve recovers the expected behavior, with temperature residuals showing only a low-level Galactic-plane feature of about $(I_1,\dots,I_N,Q,U)$5, attributed to finite sampling, while the remaining temperature and polarization residuals are consistent with noise (Galloway et al., 26 Sep 2025).

4. Antecedents in augmented CMB mapmaking

N+2 mapmaking sits within a longer line of augmented CMB inverse problems in which the map is inferred jointly with nuisance or latent variables. The Planck LFI 2015 mapmaker, implemented with Madam, uses the time-domain model

$(I_1,\dots,I_N,Q,U)$6

where $(I_1,\dots,I_N,Q,U)$7 is the sky map and $(I_1,\dots,I_N,Q,U)$8 is a vector of correlated-noise baselines. The baseline amplitudes are solved with a prior $(I_1,\dots,I_N,Q,U)$9, and the final destriped map is obtained by weighted binning of $(I,Q,U)$0. This is already an explicit example of solving for more than just the sky, with the “+k” sector represented by baseline amplitudes rather than additional sky maps (Collaboration et al., 2015).

Planck HFI 2013 mapmaking pushed the same idea into calibration. Its bogopix model writes

$(I,Q,U)$1

so that the map $(I,Q,U)$2, ring gains $(I,Q,U)$3, and ring offsets $(I,Q,U)$4 are coupled in a single calibration-aware mapmaking problem. The paper linearizes around current estimates and solves iteratively for $(I,Q,U)$5 by conjugate gradients. In the language later attached to N+2 mapmaking, this is a concrete “map plus two nuisance classes” formulation, albeit with gains and offsets rather than per-detector temperatures and shared polarization (Collaboration et al., 2013).

SRoll2 for Planck HFI generalizes the augmentation much further. Its ring-domain forward model contains the sky Stokes vector $(I,Q,U)$6 together with gains, ring offsets, transfer-function amplitudes, bandpass-mismatch coefficients, relative polarization-efficiency and angle corrections, and ADC nonlinearity spline coefficients. Rather than solving one monolithic linear system, SRoll2 alternates between nuisance-parameter estimation from data redundancy and updated map projection. It is therefore best viewed as a systematic-aware augmented mapmaker in which the sky is only one block of a much larger inverse problem (Delouis et al., 2019).

The messenger-field CMB mapmaker supplies a solver-level analogue. It rewrites the usual minimum-variance system with a latent timestream messenger field $(I,Q,U)$7,

$(I,Q,U)$8

and iterates conditional means for $(I,Q,U)$9 and $N$0. Appendices extend the construction to multiple messenger fields $N$1 for composite pointing and to $N$2 for composite noise models. This paper explicitly frames the method as an augmented-variable mapmaking architecture and identifies extra latent fields as the strongest conceptual link to generalized or “N+2” mapmaking ideas (Huffenberger et al., 2017).

5. Reduced-mode and regularized variants

A different but closely related meaning of N+2 mapmaking appears in reduced-mode inverse problems, where the unknown sky is not a full-resolution image but a deliberately small basis expansion. “Low multipole mapmaking for global 21-cm experiments” is an explicit example. There the data from multiple globally distributed zenith-pointing radiometers are modeled as

$N$3

and the reconstruction is truncated to low spherical harmonics $N$4 with $N$5. The generalized least-squares estimator

$N$6

therefore solves for the monopole plus a controlled number of low-order modes. For $N$7, the paper retains $N$8 coefficients per frequency. Unresolved higher-order modes are not ignored; they are modeled through a mean subtraction $N$9 and an additive covariance $+2$0. In the paper’s own characterization, this is a direct formalism for a reduced-mode or “N+2” mapmaking problem (Ignatov et al., 26 Jun 2025).

Regularized deconvolution for anisotropic stochastic gravitational-wave background mapmaking offers another analogue. In that setting, the dirty map obeys

$+2$1

and the maximum-posterior estimator with quadratic regularization becomes

$+2$2

The paper explicitly states that it does not present an N+2 framework, but it identifies the regularized augmented normal equation as the closest corresponding formalism if one intends “N+2” to mean adding constrained extra structure to stabilize map reconstruction (Panda et al., 2019).

Brute-force interferometric mapmaking with compact arrays provides a related linear-algebraic picture. Its core estimator,

$+2$3

is standard regularized generalized least squares, but the paper emphasizes that adding a prior map is equivalent to appending that prior to the data vector and appending an identity block to $+2$4. It does not describe this as N+2 mapmaking, yet it explicitly identifies the augmented linear-system view of priors as the closest analogy to adding extra constraints or nuisance components to a main sky solve (Zheng et al., 2016).

6. Broader uses of the “$+2$5” motif

Outside inverse problems for sky maps, the literature associated with N+2 mapmaking contains other, more metaphorical or structural uses of the “$+2$6” idea. In algebraic geometry, “Constructions of $+2$7-regular maps using finite local schemes” studies the minimum target dimension for $+2$8-regular maps $+2$9. Its refined areole/Hilbert-scheme argument improves the general upper bound from $Q$0 to $Q$1, and for $Q$2 the improvement is exactly two: $Q$3 The paper therefore singles out the $Q$4-regular case as the clearest codimension-2 phenomenon relevant to “$Q$5” mapmaking, even though the phrase itself does not occur there (Buczyński et al., 2015).

In random-matrix map enumeration, “Enumeration of maps with the Dumitriu-Edelman model” identifies the leading large-$Q$6 cumulant scaling as

$Q$7

with the $Q$8 arising from the planar Euler characteristic $Q$9. There the “$U$0” shift is not an augmented inverse problem but a topological normalization: the leading coefficient counts planar maps, while the first $U$1 correction is related to maps on $U$2 (Buc-d'Alché, 8 Dec 2025).

Combinatorial generation of maps on surfaces provides still another angle. “Generating maps on surfaces” does not give a one-shot $U$3 theorem, but it decomposes all maps on a fixed surface into irreducible triangulations, irreducible maps, face-irreducible maps, and final maps via local $U$4-expansions and $U$5-expansions. Since $U$6-expansion increases the vertex count by one, two successive $U$7-expansions realize an $U$8 growth step inside a complete recursive generation framework (Sulanke, 2015).

The cross-disciplinary spread extends further. “Generalization of Cantor Pairing Polynomials” gives an explicit bridge from classical 2D pairing to a cubic bijection $U$9, effectively moving from pairing to higher-dimensional tupling via shell-by-shell enumeration of hyperplanes $k$0. “AI Policy Projector” uses iterative mapmaking as a design process over LLM behavior space, with cases, concepts, and policies as successive layers; its core practical contribution is to let policy designers repeatedly add concepts that are present in data but not captured by the existing concept set, which is structurally close to an N+2-style expansion of the current map (Kristyan, 2024, Lam et al., 2024).

7. Limitations and open directions

The central limitation of N+2 mapmaking in the strict polarized-CMB sense is that enlarging the state vector does not remove scan-induced degeneracies. It redistributes them. Detector-specific temperature maps protect $k$1 against bandpass-mismatch leakage only when the polarization-angle distribution and cross-linking are good enough that temperature and polarization subspaces remain distinguishable. The Planck LFI 30 GHz analysis shows that detector-level solving can fail on conditioning grounds even when the underlying idea is sound (Galloway et al., 26 Sep 2025).

This limitation connects directly to older augmented mapmakers. Extra degrees of freedom require priors, regularization, or continuation strategies. Planck destriping uses a baseline prior; the HFI 2013 gain-and-offset solve is iterative and linearized; SRoll2 alternates nuisance estimation and map projection; messenger-field mapmaking depends on a cooling schedule; SGWB regularized deconvolution depends on the choice of $k$2. Across these frameworks, enlarging the unknown space is advantageous only when the additional block is either well constrained by redundancy or stabilized by explicit prior structure (Collaboration et al., 2015, Huffenberger et al., 2017, Panda et al., 2019).

A second open direction concerns granularity. The detector basis of the 2025 N+2 mapmaker is only one possible temperature decomposition. The same paper explicitly suggests horn maps as a potentially optimal compromise for Planck-like surveys, and it points toward future iterative conjugate-gradient implementations that can incorporate more realistic transfer-function structure. Reduced-mode harmonic mapmaking points in another direction: the scientifically relevant “extra” variables may be low-order sky modes rather than detector temperatures. The broader literature therefore suggests that N+2 mapmaking is less a single algorithm than a family of augmented-state designs whose success depends on choosing the right additional degrees of freedom for the instrument, scan strategy, and scientific target (Galloway et al., 26 Sep 2025, Ignatov et al., 26 Jun 2025).