1.1.

Galaxy Evolution Probe in Context

Tracing the mass assembly history of galaxies is an essential component of understanding the origins of the Hubble sequence and of using galaxies as cosmological probes of dark matter and dark energy. Observational and theoretical work over the past three decades have established a baseline framework for galaxy assembly in a cosmological context. These studies have shown that stellar and black hole mass assembly varies substantially with redshift, increasing by more than an order of magnitude between the local universe and $z=2$,^1–3 though the behavior at higher redshifts is uncertain.⁴ Environment and large-scale structure profoundly affect galaxy assembly, with factors such as local galaxy density and dark matter halo properties known to play key roles in shaping the galaxy mass function.^5–8 Finally, there is a complex and subtle relationship between star formation and active galactic nucleus (AGN) activity,⁹ including both positive and negative feedback effects.^10–13 Infrared surveys from facilities including IRAS,¹⁴ ISO,¹⁵ AKARI,¹⁶ Herschel,¹⁷ WISE,¹⁸ and Spitzer¹⁹ have played fundamental roles in these studies, since a substantial fraction of all galaxies over the history of the universe were permeated with dust through their most active evolutionary stages.

Building upon decades of observational and theoretical work, collectively a data-driven, self-consistent model for galaxy evolution that starts from cosmology, incorporates stellar dynamics and evolution, and includes interstellar processes, star formation, and supermassive black hole growth is within reach. Achieving this goal requires addressing the questions that have arisen from previous and current generations of multiwavelength surveys. These include: what was the role of feedback from black holes and stars themselves in regulating star-formation? How did galaxies’ external environments and internal contents influence their evolutionary trajectories? When were the Universe’s heavy elements formed by stars in galaxies and how did they escape into the circumgalactic and intergalactic medium?

Answering these questions will require large panchromatic surveys measuring bulk properties of hundreds of thousands to millions of galaxies over most of cosmic history paired with detailed high angular and spectral resolution studies of gas and star formation in individual galaxies. Such large samples are needed to precisely disentangle the effects of redshift and environment in driving galaxy assembly. High-resolution observations of representative and outlier galaxies identified in surveys trace detailed physical processes on a galaxy-by-galaxy basis, enabling astrophysical processes to be mapped onto cosmological processes.

New mid- and far-infrared observations enabled by the rapid advances in infrared detectors and technology²⁰ are a vital component of these next-generation surveys. They require sensitivity to detect Milky Way-type galaxies at $z=2$ ($L^{*}\sim {10}^{12}{L}_{\odot }$), prior to when most stellar mass had been assembled. Star formation rates and supermassive black hole accretion rates should be measured over a full range of cosmic environments, from isolated field galaxies, to galaxies in groups and in massive clusters. Measuring the heating of dust and gas by star formation and AGN is needed to understand how stellar and supermassive black hole growth were linked over cosmic time. Spectral mapping capability is needed for unbiased spatial–spectral surveys, line luminosities and intensity mapping, and line mapping of nearby galaxies to measure gas column densities, ionization parameters, and metallicities with tracers unaffected by dust obscuration. Crucially, galaxies detected by their dust continuum emission must have measured redshifts so their epochs and luminosity distances are known and their multiwavelength counterparts can be identified.

The Galaxy Evolution Probe (GEP) is a NASA Astrophysics Probe concept that capitalizes on new detector capability to address these questions with a powerful mid- and far-infrared toolset. GEP will measure star-formation rates and detect AGN even under conditions of heavy dust extinction. It will measure supermassive black hole accretion rates to address the connection between the masses of stellar populations and supermassive black holes. The same observations will measure metallicities with extinction-free tracers to observe growth of metals over the last two-third of the Universe’s age. In nearby galaxies, GEP will observe feedback between star-formation, AGN, and the interstellar medium to understand the processes that regulate star-formation. Mapping nearby galaxies and the Galactic interstellar medium will reveal the energy balance by measuring the total interstellar material mass, ionization state, and the local radiation field using fine-structure transitions of ions, polycyclic aromatic hydrocarbon (PAH) molecules, and the mid-infrared dust continuum.

1.2.

GEP Concept

GEP (Fig. 1) is designed with a 2.0-m primary mirror that will be cooled to 6 K to enable sensitivity limited by photon shot noise from foreground astrophysical sources: zodiacal dust emission and Galactic dust emission. There is one scientific instrument with two modules: an imager, GEP-I, and a dispersive spectrometer, GEP-S. GEP-I has 23 photometric bands distributed on the focal plane: 18 resolution $R=\lambda /\mathrm{\Delta }\lambda =8$ bands from 10 to $95\mu \mathrm{m}$ designed to measure redshifts with PAHs and mid-infrared spectral energy distributions, and five resolution $R=3.5$ bands from 95 to $400\mu \mathrm{m}$ to measure dust spectral energy distributions encompassing the peak to beyond $z=2$. GEP-S utilizes four long-slit grating spectrometers with spectral resolution $R=200$ from 24 to $193\mu \mathrm{m}$. Both modules baseline arrays of kinetic inductance detectors (KIDs) cooled to 100 mK by a multistage adiabatic demagnetization refrigerator (ADR) backed by a hybrid Joule–Thomson and Stirling cryocooler, which will also cool the telescope and coupling optics. GEP launch is targeted for January 1, 2029, with a planned mission duration of 4 years at Earth–Sun L2 (Table 1).

Fig. 1

Engineering rendering of GEP. From top to bottom, the prominent visible structures are the 2.0-m primary mirror and support structure, the sunshields, the bus with radiative panels, the sunward-facing solar-panel skirt, and the high-gain antenna. Figure reprinted from Fig. 3 of Ref. 23, with permission.

Table 1

Basic GEP parameters.

GEP is optimized for large, multitiered surveys for galaxies detected by their mid- and far-infrared emission from dust, PAHs, and atomic fine-structure lines. The GEP reference mission includes two types of surveys: photometric hyperspectral surveys with GEP-I and spectroscopic surveys with GEP-S. The GEP-I survey areas will be 3, 30, and 300 square degrees and an all-sky survey (Fig. 2). The spectral surveys with $<1\%$-level redshift precision will cover a range of low- and high-ionization atomic fine-structure lines. Spectral surveys will consist of “blind” surveys utilizing a long-slit configuration, intensity mapping, and follow-up, deep pointed observations of galaxies identified in the GEP-I surveys, and regions of the Milky Way and nearby galaxies. GEP will require 4 years to do these surveys; however, it utilizes no expendable cryogens, and a Guest Observer mission can be envisioned after the surveys are completed.

Fig. 2

GEP will detect ${10}^{12}{L}_{\odot }$ galaxies at $z=2$ and will measure redshifts with PAHs to at least $z=4$ for bright galaxies and $z\approx 7$ for gravitationally lensed galaxies. The spectra display PAH emission lines, silicate absorption at $10\mu \mathrm{m}$ (in the rest frame), a rising mid-infrared continuum from warm dust, and a peak just longward of $100\mu \mathrm{m}$ from cold dust. The spectra are binned into GEP-I’s wavebands, whose edges are demarcated by dashed vertical lines. The bandwidths change abruptly at $95\mu \mathrm{m}$ from $R=8$ to $R=3.5$ because the broad PAH emission lines are not expected to be bright enough for redshift determination for large numbers of galaxies. Atomic fine-structure emission lines and molecular lines are not shown. Magenta lines indicate $5\sigma$ survey depths assuming photon background-limited sensitivities.

GEP shares common elements with two other cryogenic far-infrared space missions under study: SPICA, an ESA-JAXA collaboration in a similar cost class to GEP, and Origins Space Telescope (OST), a NASA concept study for a future flagship infrared mission. SPICA’s design features a 2.5-m, 6- to 8-K telescope and a spectrometer using sensitive bolometer arrays, although it does not have the 23-band, moderate spectral resolution hyperspectral imaging that is central to GEP’s architecture. OST is designed with a 5.9-m, 4.5-K telescope; its earliest possible realization is well into the 2030 decade.

1.3.

This Paper

GEP was described in the concept study report submitted to NASA,²¹ and its science case was briefly summarized in a short white paper submitted to the 2020 Astronomy Decadal Survey.²² The initial design concept for the mission and the optical design were described by Glenn et al.,²³ and the cryogenic design was described by Moore et al.²⁴ This paper summarizes the design and science and provides new content: new engineering design, a summary of recent mid-infrared KID development, a discussion of microlens array fabrication for mid-infrared KIDs, and additional context for galaxy surveys.

4.1.

Optical Design

The fundamental optical design requirement is to collect light with a 2.0-m diameter primary mirror and to form an $f/9$ focus. An unobscured three-mirror astigmat (TMA) is an oft-used configuration that is well suited to the first-order optical and mechanical requirements of this system. The powered mirrors are all conic shapes with parent surfaces that have mutual tilts and decentrations to reduce wavefront error. The baseline for the GEP primary, secondary, and tertiary and chopping mirrors is unactuated silicon carbide (SiC), as was flown on the Herschel Space Observatory and Gaia. GEP’s optical design and simulated performance are described in detail in Ref. 23.

The five instrument interface planes are nearly coplanar at the common focal surface of the TMA, with minor differences in final focus to minimize wavefront error in each channel. Based on the sizes of the focal-plane array and the spectrometer slits and enclosures, the optimized field of view is $0.81\mathrm{deg}\times 0.88\mathrm{deg}$, in excess of what is required for the instrument modules. The centers of the GEP-S slits are in the plane of symmetry of the TMA so that they can be untilted with respect to the central ray to each spectrometer. Stray light suppression is achieved with a pupil stop and a field stop.

To meet the extragalactic confusion requirement (Sec. 5.3), the primary mirror is required to be diffraction limited at $\lambda =24\mu \mathrm{m}$. However, the root-mean-square (RMS) wavefront error across the field of view in the optical design is diffraction limited at the minimum wavelength of $10\mu \mathrm{m}$. The 100% encircled energy diameter is $\le {1.45}^{″}$, much less than the 3.43″ detector pixel size (see Sec. 4.2).

4.2.

GEP-I Hyperspectral Imager

GEP-I is designed to obtain repeated measurements in each of the 23 wavebands to build survey depth by continuously scanning areas of sky covered by each survey. Each waveband occupies the same amount of focal plane area, 0.002 square degrees, half of which is occupied by detectors and half of which is allotted for bandpass filter mounting in the current design configuration (Figs. 5). Bands 1 to 18 have spectral resolution $R=\lambda /\mathrm{\Delta }\lambda =8$, whereas bands 19 to 23 are $R=3.5$. It is likely that the mid-infrared (e.g., 10 to $26\mu \mathrm{m}$) bandpass filters could be replaced with linear-variable filters, resulting a spectral resolution of at least $R=20$, which will be the subject of a future investigation.

Fig. 5

GEP-I focal plane layout configuration. While the entire focal plane shown is diffraction limited at $\lambda =10\mu \mathrm{m}$, the shortest wavebands are located at the center where the optical quality is the highest. Bands 1 (10.0 to $11.3\mu \mathrm{m}$) through 15 (57.7 to $65.4\mu \mathrm{m}$) have 1440 KIDs (arrays of $12\times 120$), with 3.43″ pixels ($300\mu \mathrm{m}$ square). The longer-wavelength bands have fewer, larger KIDs such that the focal plane area occupied by each band is approximately the same. For example, band 23 (300 to $400\mu \mathrm{m}$) has 46 KIDs $1560\mu \mathrm{m}$ square.

GEP-I will have 25,735 KIDs. GEP I’s optical design performance is diffraction limited at $10\mu \mathrm{m}$; however, the primary mirror is specified to be diffraction limited at $24\mu \mathrm{m}$, corresponding to $\sim 3^{″}$ (FWHM) beam size. The primary reason for this is that the smallest KID pixel size we expect to be able to fabricate without exceeding the readout bandwidth is $300\mu \mathrm{m}\times 300\mu \mathrm{m}$ (the number of detectors is inversely proportional to their physical area), which corresponds to 3.43″. With 3″ beams and 3.43″ pixels, the full angular resolution of the telescope will not be taken advantage of for GEP-I bands 1 to 9 (up to $29\mu \mathrm{m}$), although all flux density will be recovered. For bands 9 to 23, the pixel size will be less than the beam FWHM. Should greater bandwidth become feasible through improvements in data acquisition and computing speed (a very likely development), the pixel sizes can be reduced to less than the beam FWHM.

4.3.

GEP-S Long-Slit Spectrometer

GEP’s spectrometer was designed to meet the science requirements calling for observing mid- and far-infrared atomic fine-structure lines from galaxies over a range of redshifts. Specifically, the $24.3\text{-}\mu \mathrm{m}$ [Ne V] line starting at $z=0$ for AGN identification and the $63.2\mu \mathrm{m}$ [O I] line at $z=2$ for probing star-forming galaxies at or beyond the cosmic peak. The entire bandwidth should be available to identify spectral lines for galaxies of unknown redshift. Sufficient spectral resolution is required to achieve good sensitivity through dispersion of the astrophysical background photons. Spectral resolution $R=200$ meets these requirements.

GEP-S’s design is implemented with four diffraction gratings, each with fractional bandwidth of 1.6, identified as bands 1 to 4 (Table 1), with a total of 24,640 KIDs. The ray trace diagram and CAD model of band 4 is shown in Fig. 6. Each band, or diffraction grating module, is comprised of an enclosure, a slit, a collimator, a grating operated in first order, a focusing mirror, and an array of KIDs. The long slit lengths enable spectral mapping. As with GEP-I, the shortest-wavelength bands 1 and 2 are placed nearest to the center of the field of view where the optical performance is the best. A chopping mirror [Fig. 4(a)] is included for detector modulation for staring GEP-S observations.

Fig. 6

(a) Ray-trace of GEP-S’s band 4 spectrometer module. Reprinted from Fig. 6 of Ref. 23, with permission. (b) GEP-S band 4 mechanical layout viewed mirror imaged left-to-right with respect to the top figure.

4.4.

Kinetic Inductance Detectors and Readout Electronics

KIDs have been baselined for the GEP-I and GEP-S focal plane arrays. KIDs are high-quality factor superconducting resonators whose resonances vary with absorbed light through modulation of kinetic inductance by pair breaking.⁴⁴ They are in wide use in recent and imminent ground-based and suborbital optical, far-infrared, submillimeter, and millimeter-wave instruments (Fig. 7).

Fig. 7

Far-IR detector technology has made considerable progress toward larger, more sensitive arrays over the past decade. Detector sensitivity is shown versus numbers of detectors for existing (normal font) and planned (italic font) instrumentation. The text colors indicate the type of detector technology and the font sizes are used to differentiate the testing environment or intended application of the technology. GEP-I sensitivity requirements have been met by SPACEKIDs,⁴⁵ whereas GEP-S will require further sensitivity improvement for background-limited observations. Both GEP-I and GEP-S require KIDs at mid-infrared wavelengths, for which development efforts are underway.⁴⁶ Others plotted are BLAST-TNG,⁴⁷ FIFI+LS (proposed), HIRMES,⁴⁸ MAKO,^49,50 NIKA2,⁵¹ OLIMPO,⁵² OST-OSS,⁵³ PACS,⁵⁴ SAFARI-p,⁵⁵ SAFARI,⁵⁶ SCUBA2,⁵⁷ SPIRE,⁵⁸ SuperSpec,⁵⁹ STARFIRE-p,^46,60 STARFIRE,⁶¹ and TolTEC.⁶² In addition to its own scientific program, GEP fills an important role between the detector technology achievable in the next few years and possible future flagships such as OST.

Two other detector array technologies were considered for GEP: transition-edge sensors (TESs) and Si:As IBCs (impurity band conductors). KIDs are not as technologically mature as TESs but their simple fabrication and focal plane electronics make them advantageous, partially because TESs require complex hybridization with SQUID readouts. Like KIDs, TESs have not been demonstrated in the 10 to $15\mu \mathrm{m}$ wavelength regime. Si:As IBCs are technologically mature but they are unusable beyond $28\mu \mathrm{m}$, so KIDs (or TESs) would still be needed for 28 to $400\mu \mathrm{m}$. However, pixel pitches and operating temperature requirements for IBCs (7 K) and KIDs (100 mK) are very different, and the divergent optical and cryogenic requirements would effectively necessitate two instruments each for GEP-I and GEP-S, which would drive up complexity, cost, and risk. Thus, KIDs have been adopted as the baseline detector technology for all of the wavebands of GEP-I and GEP-S.

GEP-I and GEP-S will utilize backilluminated, lumped-element, microlens-coupled aluminum KIDs. Each KID is a superconducting thin-film microresonator, comprised of an absorbing meandered inductor and an interdigitated capacitor deposited on a silicon substrate. Light is concentrated onto the inductors by a microlens array. The geometry of the inductor controls the absorption characteristics as a function of wavelength and polarization. Absorption of photons changes the kinetic inductance, producing a small shift in the resonant frequency and resonance depth, which are detected in the transmitted probe signal. To enable frequency-multiplexed readout with a minimal bandwidth requirement, the interdigitated capacitors are unique for each pixel, yielding different resonant frequencies separated by several resonance line widths. Submillimeter-wavelength KIDs with sensitivity sufficient for GEP-I have been demonstrated with a different architecture.⁴⁵ NEP improvement, not just different operational wavelengths, is needed for GEP-S.

GEP KIDs are based on the MAKO design⁴⁹ and adapted to work at shorter wavelengths. The MAKO design will work down to a wavelength of $\sim 100\mu \mathrm{m}$, but below that modification will be required: the wire grid absorber linewidth used in the $350\mu \mathrm{m}$ MAKO detectors is $0.4\mu \mathrm{m}$. Scaling this design with wavelength would require prohibitively narrow lines for $\lambda <100\mu \mathrm{m}$, even if using an ultraviolet stepper for film deposition. Our approach is to interrupt the absorber line with a meander that increases the resistance per unit length (and therefore the absorption efficiency), while simultaneously introducing a distributed capacitance to compensate for the accompanying inductance to tune the impedance.⁶³ An impedance match to silicon may then be maintained with much wider and easier-to-fabricate lines.

This absorber design is shown in Fig. 8. For $10\mu \mathrm{m}$ radiation, the linewidth is 200 nm and the unit cell is $2.4\mu \mathrm{m}$ square, repeated continuously in the circular envelope of the absorber with electrical continuity meandered vertically. Simulations indicate that this geometry achieves a peak single-polarization absorptivity of just over 70% near $10\mu \mathrm{m}$ for $R_{\square }=0.2$ to 0.8 Ω, bracketing the range expected for a 40-nm Al film. Perido et al.⁶³ presented dark measurements of the KID array shown in Fig. 8. The frequency noise was dominated by two-level system noise, with typical values at 10 Hz of $S_{\mathrm{xx}}=(3-0.6)\times {10}^{-17}{\mathrm{Hz}}^{-1}$ at $T=100$ to 200 mK, consistent with the expected $T^{-1.7}$ scaling. Perido et al.⁶³ also presented a resonant dual-polarization design for which simulations yield absorption of 70% in each polarization.

Fig. 8

A photograph the absorptive portion of a $\lambda =10\mu \mathrm{m}$ KID, courtesy H. G. LeDuc, reprinted by permission of the Creative Common license (creative commons.org/licenses/by/4.0/legalcode) from Fig. 1 of Ref. 64.⁶³ (a) The inductor/absorber portion of the KID, which has a diameter of $60\mu \mathrm{m}$. The unit cell size is $2.4\mu \mathrm{m}\times 2.4\mu \mathrm{m}$. The inductor is coupled to a larger interdigitated capacitor to form a microresonator. (b) Approximately 4 (horizontally) by 3 (vertically) unit cells. This absorber absorbs the vertical sense of polarization.

The pixel-to-pixel spacing is $300\mu \mathrm{m}$, dominated in area by the resonators’ interdigitated capacitors. Efficiently coupling the inductive absorbers to free-space radiation will be achieved with microlens arrays, and large feedhorn arrays are currently impractical for short mid-infrared wavelengths because manufacturing tolerances are too small. For minimal absorption by the substrates, the arrays should be fabricated with silicon for wavelengths $>20\mu \mathrm{m}$ and germanium for 10 to $20\mu \mathrm{m}$ because of the silicon absorption features in the 14 to $16\mu \mathrm{m}$ range. Microlens arrays could either be fabricated separately and bonded to back-illuminated KID arrays⁶⁴ or etched into the wafer frontside before or after KID fabrication. There are examples of silicon microlens fabrication in the literature that could achieve the micron-level tolerances required for operation down to $10\mu \mathrm{m}$,^65–71 although those involving micromachining and laser etching would likely be too slow for arrays of ${10}^4$ detectors. A promising approach is the deposition of Fresnel “zone plate” microlens arrays on the wafer frontsides, which have already been demonstrated at $\lambda =10.6\mu \mathrm{m}$ for antenna coupling.⁷² This approach has the virtue of fabrication simplicity but it rejects 50% of radiation. If noise-equivalent powers (NEPs) must be improved by a factor of $\sqrt{2}$, three-dimensional (3-D) hemispherical lenses or Fresnel lenses could be etched on the wafer frontsides instead of depositing Fresnel zones.

For all GEP focal planes, KIDs are organized into groups of $\sim 1500$ detectors that are spread across a 0.6 to 1.6 GHz band and read out using electronics illustrated in Fig. 9. The choice of a 1.1-GHz center frequency resulted from a trade study in which smaller pixels were favored by the optical design but larger pixels reduced the readout frequency and bandwidth, and thus power dissipation. The readout electronics generate analog waveforms using RF-DACs that are transmitted to the cold focal plane, exciting all 1500 resonators. The 1-GHz bandwidth return signal from all 1500 KIDs is digitized with RF-ADCs and digitally channelized with sufficient resolution to separate the individual KID frequencies. The KID readout scheme, initially demonstrated in 2006,⁷³ has now been implemented in various forms for ground-based and balloon-borne instruments.^50,74–82

Fig. 9

Readout system for the GEP KID arrays. Electronics will be room temperature ($\sim 300\mathrm{K}$) except the amplifiers at 4 and 18 K and the KIDs themselves at 0.1 K. The power consumption is estimated at 25 W per readout circuit. GEP-I uses 23 readout channels, GEP-S bands 1 and 2 use six readout channels each, and GEP-S Bands 3 and 4 use three readout channels each.

The GEP-I and GEP-S modules will share the same set of readout electronics (Fig. 9): they have separate, nonsimultaneous observing modes. Microwave switches on the readout lines enable selection of GEP-S or GEP-I for readout. There are 24 parallel readout channels available from three RF readout boards with eight channels each. The eight-channel RF readout cards share a payload electronics chassis with a chopping mirror driver card and two clock, processor, memory, and power cards for dual-string redundancy. The total estimated power consumption for the readout electronics is 484 W.

4.5.

KID Development and Readout Outlook

For a launch date as early as January 1, 2029, technologies required for GEP must be at or above TRL 6 in 2025. Mid-infrared KIDs are currently at TRL 3, with the proof-of-concepts having been demonstrated. To reach TRL 6, they must be validated at a component level in a relevant environment and in a subsystem operational environment. In this case, the latter indicates integration with optics and readout electronics. Here, we review the current state-of-the-art and itemize the technological improvements that must be made at a component level. This will have to followed by demonstrations of sensitivity, yield, and sufficiently low cosmic ray susceptibility for arrays while read out with flight-like electronics. Of the existing demonstrations shown in Fig. 7, the SPACEKIDs results⁴⁵ are the closest to meeting the GEP requirements, which are listed in Table 2. By comparing the capability gap between GEP KID focal plane requirements and SPACEKIDs performance, specific advances required to achieve TRL 6 can be identified.

Table 2

Detector requirements for GEP versus achieved for SPACEKIDs.45 Notes: Minimum tile sizes for GEP-I/GEP-S shown; actual arrays could be multiples thereof. Tiles with 12×120=1440  pixel format are envisioned for GEP-I bands 1 to 18. GEP-S bands 1 and 2 assume arrays with 112×70 format, which could consist of tiles with 28×35=980  pixels. The dynamic range is specified for a 1-Jy calibration source, e.g., an asteroid.45,83 Techniques to mitigate electrical and optical crosstalk, and cosmic ray susceptibility, have been demonstrated.45,84,85

Fig. 10

Detector NEPs for various photon backgrounds for the zodiacal light, the Galactic thermal dust emission, and GEP telescope, and instrument parameters. This shows why a $T_{\mathrm{tele}}\le 6\mathrm{K}$ telescope is necessary: it is not possible to achieve astrophysical photon background-limited performance for wavelengths longer than $200\mu \mathrm{m}$ with $T_{\mathrm{tele}}>6\mathrm{K}$. A cooler telescope would have better GEP-I sensitivity, but it would be of limited value because the astrophysical source confusion is expected to be severe for $\lambda \ge 200\mu \mathrm{m}$. GEP-S would benefit from cooling the telescope to 4 K even at the longest wavelengths, so this possibility may be adopted later.

To date, all KID instrument readout systems use field-programmable gate arrays (FPGAs)⁸⁸ for the digital channelization. This channelization is conceptually equivalent to an FFT and sometimes is implemented as such.^74,80 First-generation KID readouts^50,75–77 typically used Virtex-5⁸⁹ hardware to process a 500-MHz bandwidth using about 50 W, across which up to 4000 KIDs could be multiplexed.⁵⁰ Similarly, the CORE mission study⁹⁰ concluded that a 1-GHz bandwidth could be read with 50 W using existing space-grade, TRL-9 Virtex-5 FPGAs.⁹¹ Newer-generation FPGAs could have $10\times$ lower power consumption,⁹² (Fig. 11), thus the laboratory figure of 50 W per 1 GHz channel is conservative, and there is strong interest in the use of advanced FPGAs in space.^98–103 Radiation test results on late-generation FPGAs have been positive.^91,104–106 As a result, Xilinx’s 20-nm KU060 FPGA (used in the SMURF readout electronics developed at SLAC)⁸² will be available as a space-rated product by late 2020.⁹¹ This bodes well for even more advanced options, such as the new 16-nm FinFET Xilinx RF-system on chip (RFSoC), which integrates eight ADCs, eight DACs, and considerable FPGA logic into a single chip; the entire GEP readout could potentially be reduced to one or two such chips. A board with this chip has been released;¹⁰⁷ initial power dissipation estimates are well below the GEP conservative assumption of 25 W per 1 GHz channel. Alternatively, mixed-signal application-specific integrated circuits that integrate ADCs and signal processing have been developed for similar applications.¹⁰⁸

Fig. 11

As part of the GEP concept design study, a plan was developed to mature KIDs and payload electronics technologies in advance of an anticipated Probe-class proposal through laboratory (shaded in red), SOFIA and balloon-borne instruments (green), and system-level demonstrations (blue) that started in 2019. In the top of the figure, feasibility of GEP readout electronics is illustrated using the evolution of Xilinx FPGA capability over time. Specific devices are plotted as points labeled by the part number, with data sheets listed in the references. The horizontal axis gives the introduction date, and the vertical axis is a composite performance metric that combines the number of logic gates and DSP cells, memory, and IO capability. The dashed blue line indicates the approximate minimum capability required for meeting the GEP baseline power dissipation of 25 W per 1 GHz readout channel. Color is used to indicate the CMOS technology node, while font size indicates suitability for use in the space radiation environment. The XQRKU060, which Xilinx plans to release as a space-grade part in 2020, is a strong candidate. GEP’s technology development plan calls for a TRL 6 validation of all hardware required for GEP’s payload by 2025. FPGA data sheets are listed in Refs. 92–97.

5.1.

Surveys

GEP is designed to do large, unbiased surveys. Its overarching goal is to provide a legacy dataset with broad utility for studying the evolution of star-formation, interstellar medium, and black hole growth in galaxies. Parameters of GEP’s surveys in the context of other surveys are shown in Figs. 12 and 13 with complementary surveys and the astrophysics that they probe.

Fig. 12

Context for galaxy evolution surveys: a representative sample of galaxy surveys as a function of wavelength and redshift.^109–120 The emphasis is on space-based surveys and not all surveys are shown to minimize confusion in the figure. Some ground-based surveys are included that provide critical, unique information. Completed or nearly completed surveys have solid borders, while planned surveys have dashed borders. The left terminations indicate either lower redshift limits arising from observational spectral bands or are indicative of the areas surveyed. For example, WISE covered the entire sky and therefore probed very low redshifts comprehensively, whereas deep, narrow surveys such as HST CANDELS and Spitzer CANDELS probed fewer galaxies at the lowest redshifts. Terminations that fade to the right indicate reduced sensitivity to detecting galaxies at high redshifts because of flux limits.

Fig. 13

Context for galaxy evolution surveys: a representative sample of galaxy surveys as function of wavelength and sky area.^{109–112,114,115,119–126} Surveys on the left side of the panel are generally deep, optimized for detecting faint and high-redshift galaxies but are cosmic sample variance limited, whereas wide surveys toward the right generally detect brighter and lower-redshift galaxies (the exception is lensed galaxies, which are rare). GEP-I and GEP-S surveys indicated with the bold purple and blue lines occupy a broad range of wavelength and sky area space. IRAS covered the entire sky, but its observations were shallow (by comparison to GEP) and did not include spectra. GEP surveys extend sensitive mid- to far-infrared imaging and spectroscopy more than two orders of magnitude from previous surveys, finally exploring the dusty side of galaxy evolution across the whole sky and over two-third the age of the Universe.

The GEP-I survey program is optimized to sample a comprehensive range of redshifts and galaxy luminosities (Table 3). A combination of four depths and areas will sample low redshift and rare, luminous galaxies, faint, high-redshift galaxies, and intermediate redshift and luminosity galaxies. Large numbers of galaxies will be detected in each tier, preventing sample variance and small number statistics from hampering conclusions on galaxy evolution with redshift (Fig. 14). It is anticipated that the GEP-I surveys (except the all-sky survey) will be centered on and divided between the north and south ecliptic poles to minimize the photon backgrounds from primarily zodiacal dust and secondarily Galactic dust. This will provide sky coverage overlap with Euclid and Nancy Grace Roman Space Telescope (NGRST) surveys, which will provide near-infrared photometry, morphologies, and stellar masses for galaxies detected by GEP.

Table 3

GEP galaxy surveys and yields obtainable with background-limited sensitivities.

Fig. 14

Cosmic sample variance and the Poisson limit for GEP surveys. The red horizontal lines denote the depths of three of the four survey layers (excluding the all-sky survey). GEP’s “wedding-cake” survey layers observe sufficiently large angular areas that they will not be limited by cosmic sample variance in the target redshift ranges of each tier. The 3- and 30-square degree surveys will also not be limited by shot noise for $0.5\le z\le 3.5$. The 300 square degree has a lower shot noise limit ($z\sim 0.1$) and the all-sky survey will extend to even lower redshifts. The blue line shows the minimum areas needed to detect enough galaxies to enable luminosity function slope measurements to a precision of 10% at the faint end.

Three types of spectroscopic measurements with GEP-S will complement GEP-I photometric surveys: (1) individual observations of specific galaxies identified in the GEP-I surveys to provide precise redshifts and to validate PAH redshifts and to obtain measurements of the far-infrared emission lines. (2) Unbiased spectroscopic surveys obtained by rastering GEP-S on the sky. (3) Spectral maps of nearby galaxies. GEP-S will perform a deep spectroscopic survey over 1.5 square degrees and a wide spectroscopic survey over 100 square degrees (Table 3). The spectral survey datasets will detect galaxies the far-infrared fine-structure transitions (and the continuum, when binned). The wide survey will be used to stack GEP spectra on the NGRST and/or Euclid grism sources to provide high signal-to-noise ratio average galaxy spectra in bins (Table 4). The GEP-S spectral mapping speed is shown in Fig. 15, where it is seen that GEP would be a substantial improvement of the state of the art with times to survey sky regions in the far-infrared six orders of magnitude faster than previous observatories.

Table 4

GEP stacking of IR galaxy datasets obtainable with background-limited sensitivities.

Fig. 15

Spectral survey time to a given depth in the mid- and far-IR (lower is faster). The GEP-S spectrometer modules offer gains of six orders of magnitude in observing speed relative to state-of-the-art—Herschel and SOFIA—because of their warm apertures. The GEP-S speed exceeds that of SPICA because of the larger focal plane array format that enables long slits.

5.2.

Theoretical Framework and Simulations

To quantify the science yield of the GEP-I survey program, a mock survey was constructed using a combination of the Millennium N-body simulation¹²⁷ to provide the distribution of large-scale structure and the Galacticus semianalytic model¹²⁸ to populate that simulation with galaxies based on physical models. For each dark matter halo in the simulation volume, the star-formation rate and black hole accretion rate of the galaxy were computed. Bolometric infrared luminosities from star-formation were estimated as $L_{\mathrm{IR}}=2.6\times {10}^{45}$ ($\mathrm{SFR}/M_{\odot }{\mathrm{yr}}^{-1}$) ergs ${\mathrm{s}}^{-1}$,¹²⁹ and those due to AGN activity as $ϵ{\dot{M}}_{•}{\mathrm{c}}^2$ (where $ϵ$ is the radiative efficiency computed by Galacticus from the black hole spin and accretion rate, and ${\dot{M}}_{•}$ is the black hole accretion rate). Infrared spectra with the corresponding AGN fraction were then assigned to galaxies using the models of Dale et al.²⁸ and normalized to give the computed total infrared bolometric luminosity. In the Dale et al.²⁸ models, the distribution of dust heating intensities (and, therefore, temperatures) is parameterized in terms of ${\alpha }_{\mathrm{SF}}$, the exponent of the power-law distribution of heating intensities. Values of ${\alpha }_{\mathrm{SF}}$ for model galaxies were drawn from a Gaussian distribution with mean, ${\overline{\alpha }}_{\mathrm{SF}}=1.75$, and dispersion, ${\sigma }_{{\alpha }_{\mathrm{SF}}}=0.25$, with no redshift dependence. A range of different possible values of ${\overline{\alpha }}_{\mathrm{SF}}$ and ${\sigma }_{{\alpha }_{\mathrm{SF}}}$ was explored, with the above values selected as those which result in the closest match between our simulated and observed numbers counts of galaxies over the 24- and $350\text{-}\mu \mathrm{m}$ wavelength range.^130,131 These spectra were then used to compute broadband luminosities for model galaxies in each GEP-I band. Finally, a light cone from this synthetic catalog was extracted, corresponding to a 4-square degree area, from $z=0$ to 3, and observed fluxes were determined in all GEP-I bands for each galaxy.

The galaxy number counts estimated in this way are lower limits because these (and other) galaxy evolution models generically underpredict the number of very high luminosity galaxies¹³² ($\sim {10}^{13}{L}_{\odot }$). Although we attempted to tune the Galacticus model number counts to match observations from Spitzer and Herschel by judiciously choosing from the Dale et al.²⁸ spectral library, at high flux densities (i.e., $>1\mathrm{mJy}$ at $160\mu \mathrm{m}$) the model counts are almost an order of magnitude too low compared with observations.¹³¹ The agreement is much better at 24 and $70\mu \mathrm{m}$, coming close to matching observations.¹³⁰ Some of the bright observed galaxies that are not accounted for in the models likely result from gravitational lensing, although lensing is unlikely to account for the majority of the discrepancy. We adopt our model predictions for the detection rates with the understanding that they are likely conservative.

Noise was added to the simulated maps corresponding to the expected map depths after they were convolved with GEP beam sizes. A small region of the maps stepping through the bands is animated in Fig. 16.

Fig. 16

A $10\text{-}\mu \mathrm{m}$ still image from Video 1, a rendering of the Galacticus simulations. The axis units are pixel number, with ${3.43}^{″}\times {3.43}^{″}\text{pixels}$—the image is 17.2′ on a side. The animation runs continuously through the bands from 10 to $80\mu \mathrm{m}$ showing that galaxies alternately brighten and dim as PAH emission lines move through the bands and can thus be used to measure redshifts (Video 1, MP4, 3.5 MB [URL: https://doi.org/10.1117/1.JATIS.7.3.034004.1]).

5.3.

Extragalactic Source Confusion

Extragalactic source confusion arises when images of galaxies—the point spread function for unresolved galaxies—overlap. It happens at far-infrared wavelengths where diffraction-limited beams can be several arcseconds or larger. Confusion “noise” is the uncertainty in the extraction of a given source’s flux due to the presence of a background of other sources which cannot accurately be subtracted from the signal. Analytically, the confusion noise ($1\sigma$) is just the standard deviation of the flux from beam to beam, and for extraction of sources from a map, a “confusion limit” arises at $3\text{–}5\times$ this $1\sigma$ value. This limiting flux below which extractions are unreliable typically corresponds to that for which density of sources is 1 per several ($\sim 10$) beams, depending on the source count relation. This basic paradigm holds true when referring to extracting sources from a map without prior information, but there are well-developed methods to employ prior higher-resolution datasets to extract source parameters in a larger far-IR beam. An excellent example is use of Spitzer IRAC and MIPS $24\mu \mathrm{m}$ positions to extract (statistically) fluxes in the 70 and $160\mu \mathrm{m}$ maps.¹³³ In this section, we assess the important issue of source confusion with the GEP-I datasets, but first note that spectral surveys with GEP-S should not suffer appreciably from source confusion.

For spectral mapping with GEP-S, source confusion should not limit the extraction of line intensities. The key point is that the bright mid- and far-IR spectral lines are sparse and well-separated spectrally, and they form an unambiguous template for redshift identification. Thus, each line can be conclusively identified and measured, even with multiple line emitting sources in the beam. For a 3-D spatial–spectral survey, the analog of source counts for two-dimensional is line counts, the number of line emitters above a given flux level in a typical spatial–spectral bin of the survey. This issue was studied for the OST, which also features a wideband, moderate-resolving-power spectrograph; Ref. 134 is a good reference. As an example they predict that at $190\mu \mathrm{m}$ (which is approximately GEP-S long-wavelength range limit), the density of lines with fluxes above ${10}^{-20}\mathrm{W}{\mathrm{m}}^{-2}$ (corresponding to a 25-h integration with GEP-S) per beam-bin is 1/150 for Origins. Correcting this for the larger GEP beam ${(2/5.9)}^{-2}$ and slightly smaller resolving power ${(200/300)}^{-1}$ results in an estimate of one source per 12 beam-bins for GEP-S at this wavelength. This is approaching but not exceeding a practical confusion limit. Shorter wavelengths of course are more forgiving with the smaller beam.

We estimate the extragalactic confusion noise to be expected for GEP-I observations by considering confusion noise measurements from previous observations. The confusion noise flux density was measured—or upper limits were placed—with Spitzer 70 and $160\mu \mathrm{m}$^135,136 and Herschel 70, 100, and $250\mu \mathrm{m}$.^131,137,138 We start by deriving an empirical relationship for confusion noise as a function of telescope aperture diameter to assess the dependence of confusion on telescope size and to check for consistency between Spitzer and Herschel observations, which had different wavebands and aperture sizes (0.85 and 3.5 m, respectively). Using the number counts models of Béthermin et al.,¹³⁹ the $\lambda =70\mu \mathrm{m}$ flux density corresponding to 15 beams per source (where galaxies are not be substantially blended) scales approximately as the inverse of aperture diameter squared ($\propto D^{-2}$, slightly steeper for apertures smaller than 1.7 m, Fig. 17). Confusion noise measurements in the literature cited above scaled by this relationship are consistent, which validates the scaling relation for interpolation to the GEP 2.0 m aperture for estimation of confusion noise. Critically, the scaling also shows that aperture diameters below 2.0 m will be increasingly susceptible to confusion. Since the confusion flux density drops slowly with telescope diameter beyond 2.0 m, 2.0 m represents something of a “sweet spot” in the trade between confusion and cost.

Fig. 17

The flux density for 15 beams per galaxy as a function of telescope aperture diameter at $\lambda =70\mu \mathrm{m}$. The number counts are from Ref. 139 and the curves are piecewise fits. GEP’s 2.0 m aperture is a “sweet spot” between rapidly worsening confusion for smaller apertures and increasing cost for larger apertures.

A comparison of the expected GEP-I survey map depths for a 2.0-m aperture with the confusion measurements cited in the literature already mentioned yields the following:

In short, GEP-I will likely have significant confusion noise at $\lambda =70\mu \mathrm{m}$ and longer, but not at shorter wavelengths. GEP-I will have to integrate deeper than the $70\mu \mathrm{m}$ confusion noise in the two deepest surveys for PAH redshifts of $z\le 4$ galaxies with the wavebands at $50\mu \mathrm{m}$ and below, which will not be limited by confusion noise. In addition, monochromatic fluctuation “probability of deflection” P(D) analyses show that it is possible to constrain galaxy populations meaningfully with observations deeper than the confusion noise level.^140,141 A polychromatic P(D) analysis with GEP-I observations covering the redshifted PAH features would be extremely powerful; it would yield precise galaxy number counts and redshifts statistically for the ensemble, and therefore luminosity functions, and tightly constrain galaxy evolution models. Furthermore, using cross-identification with counterparts at shorter wavelengths, galaxy properties can be measured even when there is source confusion. For example, Labbé et al.¹⁴² showed that contamination by confusion can be reduced a factor of six with short-wavelength prior-based photometry, and positional and flux priors from ancillary data could enable improved spectral energy distribution extraction. In addition, because the detector count in wavebands longer than $100\mu \mathrm{m}$ (GEP-I band 19) is small (4.2) compared with shorter wavelengths, the cost of retaining them is merited for measurements of far-infrared luminosities, for nearby galaxy science that do not require observations as deep, and for P(D)-type fluctuation analyses.

For wavelengths $\lambda \ge 70\mu \mathrm{m}$, corresponding to GEP-I bands 17 to 23, photometry of partly blended sources can still be extracted reliably. The Next Generation (X)Cross Identification (XID+) code was developed to estimate flux densities accurately from confusion-limited Herschel imaging;¹⁴³ we apply it here to GEP-I. The performance of XID+ was quantified by measuring the differences between fitted and true galaxy flux densities as a function of the distance to the nearest galaxy neighbor using the simulations described in Sec. 5.2. Positional priors were assumed from the mid-infrared wavebands and the fitted galaxy far-infrared flux densities were compared to their true values. Figure 18 shows that down to galaxy separations of the beam size—below the classical confusion “limit”—galaxy flux densities can be deblended with small fractional errors and little or no bias in most cases. Using the deblended sources, the map flux densities were reproduced with small residuals, although with poorer performance at the map edges because of missing galaxies outside the field of view. This will not be a concern for the majority of galaxies detected in GEP-I surveys, for which the minimum size is 3 square degrees.

Fig. 18

Differences between XID+ fitted and true galaxy flux densities as a function of distance to the nearest galaxy neighbor for three GEP-I long-wavelength bands: 17, 20, and 23 (with FWHM beam sizes of 8.6″, 16″, and 38″, respectively). Red vertical lines show the beam width as measured by the full width at half maximum. For bands 17 and 20, the fractional error in the flux density is small down to the beam size and clustered around zero. At the beam size and below the dispersion increases, although there is no bias. For band 23, GEP-I’s longest-wavelength band, the simulation data do not extend to great enough separations to see that the offset decreases toward zero for separations above the beam size. In this band, there is a negative bias for some galaxies that will have to be corrected.

5.4.

Anticipated Science

5.4.1.

GEP-I Redshift Measurements

To measure the cosmic star formation history of the Universe, GEP requires redshifts with precision ${\sigma }_z/(1+z)\le 10\%$ to $z=2$, a requirement that our simulations show is achieved with margin. In general, the redshift uncertainty is set by the width of the GEP-I bands relative to the widths of the PAH features.

Expected GEP-I redshift precision was estimated by adding noise to the model spectra according to each of the map depths given by the photon backgrounds. The mock spectra were binned into the GEP-I bands and ${\chi }_{\nu }^2$ were calculated by comparison to the spectral model and two other different spectral models: since the spectra of galaxies will not be known a priori, the comparison models were used to ascertain the uncertainties incurred by having a spectrum different from the model. Only the first 11 GEP-I bands (10 to $37\mu \mathrm{m}$) were used for the redshift measurements because the steeply rising mid-infrared dust spectra influence the redshifts and the dust temperatures will not be known a priori. A sample GEP-I redshift chi-squared distribution is shown in Fig. 19. The nominal model had strong PAH emission features and the alternate comparison models had: (1) strong PAH emission features but cooler dust (hence a more slowly rising spectrum with wavelength) and (2) hot dust that substantially overwhelmed the PAH features above $10\mu \mathrm{m}$. The results were as follows:

• 1. Redshifts are obtainable for ${10}^{11}{L}_{\odot }$ galaxies (corresponding to a median stellar mass of $4\times {10}^9{\mathrm{M}}_{\odot }$) out to $z=2$ in the 3 square degree survey with ${\sigma }_z/(1+z)\le 0.1$.

• 2. Redshifts are obtainable for ${10}^{12}{L}_{\odot }$ galaxies (corresponding to a median stellar mass of $1\times {10}^{10}{\mathrm{M}}_{\odot }$) to $z=2$ in the 300 square degree survey and to $z=4$ in the 30 square degree survey with ${\sigma }_z/(1+z)\le 0.1$.

• 3. Redshifts are attainable for ${10}^{13}{L}_{\odot }$ galaxies (corresponding to stellar masses in excess of ${10}^{11}{M}_{\odot }$) to $z=4$ in the 300 square degree survey ${\sigma }_z/(1+z)\le 0.05$.

• 4. The redshift uncertainties are a function of redshift, galaxy luminosity, map depth, and strength of the PAH features relative to the continuum.

• 5. In the case of very warm dust, which represents an extreme case of entirely AGN-dominated galaxies, the redshifts have large uncertainties. However, even in this case, redshifts become detectable at $z>2$ for deep surveys and luminous galaxies because the dust continuum is not strong below rest-frame $10\mu \mathrm{m}$ and there is a strong $3\mu \mathrm{m}$ emission line, which is redshifted well into the GEP-I bands, and silicate absorption can still be present.

Fig. 19

The redshift reduced chi-squared distribution for a $z=2$, $L={10}^{12}{L}_{\odot }$ galaxy with the model spectrum shown in Fig. 2. The depth corresponds to the 30 square degree survey. The measured redshift is $z=2\pm 0.02$, with ${\sigma }_z/(1+z)=0.01$. This represents a best case because the observed spectrum was fit to the same model spectrum (same dust parameters) and ideal square $R=8$ bandpasses were assumed. Uncertainties from real GEP observations will be larger; however, this analysis shows that the multiple PAH emission lines combined with the deep silicate absorption can in principle lead to precise redshift determination. The jagged nature of the chi-squared distribution arises from the sampling of the model spectrum.

A comprehensive set of simulations is required to quantify redshift measurements as a function of GEP-I spectral resolving power, AGN fraction and other sources of weak PAH emission (e.g., low metallicity), galaxy-to-galaxy variability, and instrument parameters, such as nonideal bandpass shapes and $1/f$ noise. These simulations likely represent best-case scenarios. These effects will be explored as a subsequent effort.

5.4.3.

Galaxy Luminosity Functions, Star Formation History, Separating AGN, and Clustering

GEP’s science objectives require luminosity functions over a range of redshifts to observe the changes in galaxy formation and the build-up of stellar mass over cosmic time. Faint-end (below $L^{*}$) mid- and far-infrared luminosity function slopes have not been measured above $z=0.5$, and there is disagreement about the faint-end slopes even at $z\sim 0$. GEP-I’s surveys will detect hundreds of millions of galaxies and measure redshifts for millions of them for derivation of luminosity functions. With GEP-I surveys, faint-end slopes below ${\mathrm{Log}}_{10}$ $(L_{\mathrm{IR}}/L_{\odot })=11$ for $z=0.5$, below ${\mathrm{Log}}_{10}$ ($L_{\mathrm{IR}}/L_{\odot })=11.5$ at $z=1$, and below ${\mathrm{Log}}_{10}$ $(L_{\mathrm{IR}}/L_{\odot })=12$ at $z=2$, will be measured. Figure 19 shows current observational determinations of the infrared bolometric luminosity function and compares these to a sample luminosity function derived from our mock GEP catalogs.

One of GEP’s primary goals is to quantify the star formation history of the Universe by utilizing very large statistical samples of galaxies that overcome Poisson statistics and sample variance from large-scale structure. How well will it do? Figure 14 showed that Poisson statistics and sample variance will be overcome and Fig. 20 demonstrated that luminosity functions will be well quantified even below $L^{*}$ at $z=1$. How do these translate into measurement of the history of cosmic star formation rate density? Figure 21 shows that the star formation history will be quantified with uncertainties an order of magnitude below the state of the art to at least $z=3$. Uncertainties will likely improve substantially beyond $z=3$ also; however, that was not included in the simulations (Sec. 5.2). Because it maps large areas, GEP will also be effective at detecting gravitationally lensed galaxies. This will enable it to serendipitously detect sub-$L^{*}$ galaxies at $z=2$ with a tail out to higher redshifts, perhaps ${10}^4$ lensed galaxies in the all-sky survey and 50 in the deeper 300-square degree survey.

Fig. 20

GEP luminosity function for $1.0<z<1.2$ derived from the Galacticus mock catalogs assuming a ${\sigma }_z/(1+z)=0.1$ uncertainty on galaxy redshifts. Error bars are estimated assuming Poisson statistics scaled from the 4 square degree area of our mocks to 30 square degrees. Blue points show the luminosity function in the GEP-I band 7 (21.2 to $24.0\mu \mathrm{m}$), while green points show the bolometric infrared luminosity function. The bolometric luminosity function is not shown below ${10}^9{L}_{\odot }$ because it becomes incomplete due to the resolution of the simulation. The blue line is the $3\sigma$ detection limit. Faint points indicate the luminosity function that would be obtained if spectroscopic redshifts were available; it is almost indistinguishable from that constructed using photometric redshifts. The orange lines are from Ref. 25.

Fig. 21

A recent study of the cosmic star formation density history from deep multiband data—for which the principal source of uncertainty is the limitation of the far-infrared data—reproduced from Fig. 13 of Ref. 153 with GEP simulation data overplotted. Blue circles show expectations from a combination of GEP full sky, 300, 30, and 3 square degree surveys (for $0<z<3$ only), showing that GEP will measure the star formation history with unprecedented precision. In most cases, error bars are smaller than the symbols. The large scatter in the GEP points arises from simulations of limited cosmic volume. While the error bar sizes are appropriate for the GEP surveys, the scatter in the GEP survey data should be very small compared with what is seen here.

Another GEP goal is to identify embedded AGN and quantify their contributions to the infrared luminosities of galaxies. The mid-infrared part of the spectrum provides excellent leverage for this because the dust is hotter—thus the spectra are “bluer”—and the PAHs are more subsumed by the bright dust continuum. Figure 22 shows that a simple principal component analysis with only three components is sufficient to identify AGN and separate their contributions to infrared luminosities from star formation with GEP-I data. The high-ionization lines observed with GEP-S will indicate the presence of AGN and enable the larger catalog of GEP-I galaxies to be calibrated against the fine-structure line luminosities. These observations can be used to assess AGN influence on galactic interstellar media and the coevolution of supermassive black holes and star formation.

Fig. 22

GEP-I’s multiband observations will enable the relative contributions of galaxy infrared luminosities from AGN and star formation to be discriminated based on their different mid-and far-infrared spectra. (a) Outcome for analysis of the 3-square degree survey at redshift $z=1.0\pm 0.1$ when all GEP bands are detected. Results are similar for other survey depths and subsets of GEP bands provided adequate detection rates. (b) The three principal components identified by the algorithm. Together, they measure the AGN fraction, radiation field hardness, and the redshift.

Within a $\mathrm{\Lambda }\mathrm{CDM}$ cosmology, the clustering of galaxies reveals the masses of dark matter haloes that they occupy. Galaxy mass correlates strongly with halo mass; however, it is possible that mid- and far-infrared luminosities of galaxies correlate more weakly with halo mass because even low-mass galaxies can be temporarily infrared-bright from star formation bursts. Clustering of GEP-I catalogs will strongly test galaxy formation and evolution theories that predict halo occupations as a function of redshift (Fig. 23).

Fig. 23

Projected correlation functions from the mock catalog for galaxies selected by bolometric infrared luminosity (solid lines) and stellar mass (dashed lines) in a $1.0<z<1.2$ redshift slice. Bands indicate the expected $1\sigma$ uncertainties for the 30 square degree survey. The Galacticus simulations show what is already known: galaxy mass correlates strongly with halo mass, leading to a strong dependence of clustering strength with mass. Conversely, however, it predicts that bolometric infrared luminosity (which indicates star-formation rate) is weakly correlated with halo mass because even galaxies in low mass halos can have occasional strong starbursts, leading to a weak dependence of clustering strength on infrared luminosity.

5.4.4.

Milky Way and Nearby Galaxies

GEP will map the Milky Way and the Magellanic Clouds with GEP-I during the all-sky survey. The sensitivity required of the telescope and instruments for the deep extragalactic observations will enable high signal-to-noise ratio mid- and far-infrared spectral energy distributions of star-formation regions. The 3.43″ mid-infrared resolution will resolve gradients in dust temperatures and PAH excitation (Fig. 24), thereby probing the radiation fields and chemistry, whereas the far-infrared spectral energy distributions will yield luminosities of embedded stars, dust temperatures, and gas masses.

Fig. 24

GEP-I will provide spatially resolved spectral energy distribution mapping across the star-forming interstellar medium over ${10}^{7.5}$ lines of sight in the Milky Way and nearby galaxies, providing a large data set to understand how local environment establishes interstellar conditions. The figure shows how the dust emission from (a) the Aquila molecular cloud would be mapped into (b) spatially resolved spectral energy distributions over the GEP-I bands (band edges denoted with blue vertical lines) with flux density estimates in 3 arc sec apertures. The curves correspond to the expected signatures for varying levels of molecular gas in a stellar population with a stellar surface density of $300M_{\odot }{\mathrm{pc}}^{-2}$. The spectral slopes, fine-structure line-to-continuum ratios, and PAH feature strengths vary depending on the gas surface densities. The horizontal lines indicate the approximate $5\sigma$ depths of the all-sky survey (solid) and 300 square degree survey (dashed). The depth flux densities are lower limits because they do not account for the increased background in the galactic plane. In addition, detector nonlinearities from large backgrounds have not been modeled or accounted for.

5.5.

Future Science Exploration

GEP’s science reach will be far broader than outlined here. For example, monochromatic probability-of-density—P(D)—fluctuation analyses have shown to be effective for constraining galaxy number counts into the confusion noise.^140,141 Multicolor P(D) analyses could go much further: distinguishing between galaxy number count models and thereby constraining luminosity functions as a function of redshift to substantially greater depth than for analyses restricted to galaxies whose brightness are above individual detection thresholds. In nearby galaxies, mid- and far-infrared spectral energy distributions are needed for panchromatic spectral energy distribution fitting to constrain stellar populations, nebular conditions, star formation rates, and identify embedded AGN on a galaxy-by-galaxy basis. In the Milky Way, mapping of PAH emission in various environments, such as the vicinities of hot, young stars, and photodissociation regions, will probe the radiation fields and PAH production, excitation, and destruction. Lastly, with the possibility of linear-variable filters with $R>8$, the optimum resolution for PAH redshifts should be revisited. Then, GEP-I redshift precision should be quantified as a function of galaxy luminosity, redshift, AGN fraction (with strong AGN mid-infrared continuum leading to weak PAH lines), and instrumental parameters, such as spectral resolving power $R$.

Finally, it must be noted that the GEP design reference mission is designed for dedicated surveys. The surveys should be designed with community input and consideration of the rich landscape of imminent multiwavelength galaxy surveys. And, while the design reference mission does not support an open-time phase to manage costs, with no expendable cryogens the expected lifetime of the mission exceeds the planned survey durations. An extended mission with guest-observer opportunities would yield a large volume of science, as the Spitzer extended mission has shown.