MIT-CTP 4763, PUPT 2497

Broadband and Resonant Approaches to Axion Dark Matter Detection

Yonatan Kahn,1, ∗ Benjamin R. Safdi,2, † and Jesse Thaler2, ‡ 1Department of Physics, Princeton University, Princeton, NJ 08544, U.S.A. 2Center for Theoretical Physics, Massachusetts Institute of Technology, Cambridge, MA 02139, U.S.A. (Dated: September 29, 2016) When ultralight axion dark matter encounters a static magnetic field, it sources an effective electric current that follows the magnetic field lines and oscillates at the axion Compton frequency. We propose a new experiment to detect this axion effective current. In the presence of axion dark matter, a large toroidal magnet will act like an oscillating current ring, whose induced magnetic flux can be measured by an external pickup loop inductively coupled to a SQUID magnetometer. We consider both resonant and broadband readout circuits and show that a broadband approach has advantages at small axion masses. We estimate the reach of this design, taking into account the irreducible sources of noise, and demonstrate potential sensitivity to axion-like dark matter with masses in the range of 10−14 − 10−6 eV. In particular, both the broadband and resonant strategies can probe the QCD axion with a GUT-scale decay constant.

A broad class of well-motivated dark matter (DM) ADMX is based on resonant detection of a cavity exci- models consists of light pseudoscalar particles a coupled tation, our design is based on either broadband or reso- weakly to electromagnetism [1–3]. The most famous ex- nant detection of an oscillating magnetic flux with sen- ample is the QCD axion [4–7], which was originally pro- sitive magnetometers, sourced by an axion effective cur- posed to solve the strong CP problem. More generally, rent. Our static magnetic field is generated by a su- string compactifications often predict a large number of perconducting toroid, which has the advantage that the axion-like particles (ALPs) [8], with Planck-suppressed flux readout system can be external to the toroid, in a couplings to electric (E) and magnetic (B) fields of the region of ideally zero static field. Crucially, this setup form a E · B. Unlike QCD axions, generic ALPs do not can probe axions whose Compton wavelength is much necessarily couple to the QCD operator GG˜, where G larger than the size of the toroid. If this experiment were is the QCD field strength. The masses and couplings built, we propose the acronym ABRACADABRA, for “A of ALP DM candidates are relatively unconstrained by Broadband/Resonant Approach to Cosmic Axion Detec- theory or experiment (see Refs. [9–11] for reviews). It tion with an Amplifying B-field Ring Apparatus.” is therefore important to develop search strategies that For ultralight (sub-eV) axion DM, it is appropriate to cover many orders of magnitude in the axion parameter treat a as a coherent classical field, since large DM num- space. ber densities imply macroscopic occupation numbers for The ADMX experiment [12–14] has already placed each quantum state. Solving the classical equation of stringent constraints on axion DM in a narrow mass motion with zero DM velocity yields −6 range around ma ∼ few × 10 eV. However, ADMX is √ 2ρDM only sensitive to axion DM whose Compton wavelength a(t) = a0 sin(mat) = sin(mat) , (2) is comparable to the size of the resonant cavity. For the ma 3 1 QCD axion, the axion mass ma is related to the Peccei- where ρDM ≈ 0.3 GeV/cm is the local DM density [15]. Quinn (PQ) symmetry-breaking scale fa via Through the coupling to the QED field strength Fµν ,

fama ' fπmπ, (1) 1 µν L ⊃ − gaγγ aFµν Fe , (3) 4 arXiv:1602.01086v3 [hep-ph] 27 Sep 2016 where mπ ≈ 140 MeV (fπ ≈ 92 MeV) is the pion mass (decay constant). Lighter QCD axion masses therefore a generic axion will modify Maxwell’s equations [16], and Amp`ere’scircuit law becomes correspond to higher-scale axion decay constants fa. The 16 −9 GUT scale (fa ∼ 10 GeV, ma ∼ 10 eV) is par- ∂E  ∂a ∇ × B = − g E × ∇a − B , (4) ticularly well motivated, but well beyond the reach of ∂t aγγ ∂t ADMX as such small ma would require much larger cav- ities. More general ALPs can also have lighter masses with similar modifications to Gauss’s law. For the QCD and larger couplings than in the QCD case. axion, gaγγ = gαEM/(2πfa), where αEM is the electro- In this Letter, we propose a new experimental de- magnetic fine-structure constant and g is an O(1) num- sign for axion DM detection that targets the mass range ber equal to ∼ 0.75 (−1.92) for the DFSZ model [17, 18] −14 −6 ma ∈ [10 , 10 ] eV. Like ADMX, this design ex- ploits the fact that axion DM, in the presence of a static magnetic field, produces response electromagnetic fields 1 The local virial DM velocity v ∼ 10−3 will give small spatial that oscillate at the axion Compton frequency. Whereas gradients ∇a ∝ v. 2

ν=��/�π �� ��� ��� ���

� -�� ������� ��� = � ��� � =�� �� � ������� ��� = �� ��� � =�� � ������� ��� = � ��� � = ��� � � R ����� ��� = � ��� � =�� -�� � r �� ����� ��� = �� ��� � =�� � ����� ��� = � ��� � = ��� �

a ) -��

- � �� ( ��� h -�� � γγ �� B0 �

��-�� Figure 1. A (gapped) toroidal geometry to generate a static magnetic field B0. The dashed red circle shows the location of the superconducting pickup loop of radius r ≤ R. The gap -�� �� ���� ���� ���� ���� ensures a return path for the Meissner screening current; see �� (���) discussion in main text. ��-�� ��-�� ��-�� ��-� ��-� �� (��) (KSVZ model [19, 20]). Thus, in the presence of a static

magnetic background B0, there is an axion-sourced ef- Figure 2. Anticipated reach in the gaγγ vs. ma plane for fective current the broadband (Broad) and resonant (Res) strategies. The benchmark parameters are T = 0.1 K, r = a = R = h/3 2 (see Fig. 1), and Lp = Lmin ≈ πR /h. The total measure- p ment time for both strategies is t = 1 yr, where the resonant Jeff = gaγγ 2ρDM cos(mat)B0. (5) experiment scans from 1 Hz to 100 MHz. The expected pa- This effective current then sources a real magnetic field, rameters for the QCD axion are shown in shaded red, with oscillating at frequency m , that is perpendicular to B . the corresponding decay constant fa inset at bottom right. a 0 The projected sensitivities of IAXO [41] and ADMX [14] are Our proposed design is shown schematically in Fig. 1. shown shaded in light green. Published limits from ADMX The static magnetic field B0 is generated by a constant [13] are shown in gray. current in a superconducting wire wrapping a toroid, and the axion effective current is detected with a supercon- terference device (SQUID). The broadband circuit uses ducting pickup loop in the toroid hole. In the absence of a untuned magnetometer in an ideally zero-resistance axion DM (or noise), there is no magnetic flux through setup, while the resonant circuit uses a tuned magne- the pickup loop. With axion DM, there will be an os- tometer with irreducible resistance. Both readout cir- cillating magnetic flux through the pickup loop propor- cuits can probe multiple orders of magnitude in the axion √ tional to ρDM. This design is inspired by cryogenic cur- DM parameter space, though the broadband approach rent comparators (CCCs) [21], which are used for mea- has increased sensitivity at low axion masses. suring real currents. The key difference here is the static A related proposal, utilizing the axion effective current, external field B0, which generates an effective electric was put forth recently by Ref. [22] (see also Ref. [23] for a current in the presence of axion DM instead of the real preliminary proposal and Ref. [24] for a similar design for current in the case of the CCC. detecting dark photon DM). That design was based on In a real implementation of both designs, the signal a solenoidal magnetic field, with the pickup loop located flux is actually sourced by a Meissner current which re- inside of the solenoid, and focused on resonant readout turns along the outside surface of a gapped toroid. The using an LC circuit. The design presented here offers a size of the gap is not crucial for our analysis, but must few advantages. First, the toroidal geometry significantly be sufficiently large that parasitic capacitance effects do reduces fringe fields compared to a solenoidal geometry. not generate a displacement current, which might shunt Second, the pickup loop is located in an ideally zero- the Meissner return current and reduce the induced sig- field region, outside of the toroidal magnetic field B0, nal B-field. For wires of diameter 1 mm and a meter- which should help reduce flux noise. Third, as we will sized toroid, a gap of a few millimeters allows unscreened show, broadband readout has significant advantages over currents up to the frequency at which the magnetoqua- resonant readout at low axion masses. Our proposal is sistatic approximation breaks down and displacement complementary to the recently proposed CASPEr exper- currents are unavoidable. In what follows, we will es- iment [25], which probes a similar range of axion masses timate our sensitivity using the axion effective current but measures the coupling to nuclear electric dipole mo- which is correct up to O(1) geometric factors. ments rather than the coupling to QED. See Refs. [26–40] We consider two distinct circuits for reading out the for other proposals to detect axion DM. signal, both based on a superconducting quantum in- For concreteness, our sensitivity studies are based on 3

Lp [42]: M Δω M s α L Li Lp L L ΦSQUID ≈ Φpickup. (9) p C L 2 Lp Li R Here α is an O(1) number, with α2 ≈ 0.5 in typical Figure 3. Schematics of our readout circuits. Left: broad- SQUID geometries [43]. band (untuned magnetometer). The pickup loop Lp is placed Clearly, the flux through the SQUID will be maximized in the toroid hole as in Fig. 1 and connected in series with for L as large as possible and Lp as small as possible. A an input coil Li, which has mutual inductance M with the typical SQUID has inductance L = 1 nH. A supercon- SQUID of self-inductance L. Right: resonant (tuned mag- ducting pickup loop of wire radius φ = 1 mm and loop netometer). L is now in series with both L and a tun- p i radius r = 0.85 m has geometric inductance of [44] able capacitor C. A “black box” feedback circuit modulates the bandwidth ∆ω and has mutual inductance M with the SQUID. Lp = r(ln(8r/φ) − 2) ≈ 7 µH, (10) but this may be reduced with smaller loops in parallel a toroid of rectangular cross section (height h, width a) as in a fractional-turn magnetometer [45, 46]. The mini- and inner radius R, as illustrated in Fig. 1. The magnetic mum inductance is limited by the magnetic field energy 1 R 2 field inside the toroid volume is 2 B dV stored in the axion-sourced response field, and is approximately R ˆ (6) B0(s) = Bmax φ, 2 s Lmin ≈ πR /h. (11) where s is the distance from the central axis of the toroid, With a “tall” toroid where h = 3R, one can achieve φˆ is the azimuthal direction, and B is the magnitude max Lmin ≈ 1 µH and ΦSQUID ≈ 0.01Φpickup for R = 0.85 of B0 at the inner radius. The flux through the pickup m. Since the pickup loop area is much larger than the loop of radius r ≤ R can be written as magnetometer area, the B-field felt by the SQUID is sig- p nificantly enhanced compared to the axion-induced field Φpickup(t) = gaγγ Bmax 2ρDM cos(mat) VB. (7) in the pickup loop. The B-field enhancement takes ad- vantage of the fact that we are working in the near-field The effective volume containing the external B-field is limit, so that the induced B-field adds coherently over the pickup loop. Z r Z R+a Z 2π 0 0 0 Rhr (s − r cos θ) To assess the sensitivity of the untuned magnetome- VB = dr ds dθ √ , (8) 2 2 2 0 R 0 r˜ h + 4˜r ter to the axion-sourced oscillating flux in (7), we must characterize the noise of the circuit. In a pure supercon- 2 2 02 0 withr ˜ ≡ s + r − 2sr cos θ. We work in the magneto- ducting circuit at low frequencies, there is zero noise in quasistatic limit, 2π/ma  r, R, h, a; at higher frequen- the pickup loop and input coil, and the only source of cies, displacement currents can potentially screen our sig- noise is in the SQUID, with contributions from thermal nal. As an illustration, we consider a meter-sized exper- fluctuations of both voltage and current. Despite their 3 iment, where VB = 1 m for r = R = a = h/3 = 0.85 thermal origin, we will refer to these as “magnetome- < −6 m, with sensitivity to ma ∼ 10 eV. For an example of ter noise” to distinguish them from noise in the pickup the magnitude of the generated fields, the average B-field loop circuit (which dominates in the resonant case be- 16 sourced by a GUT-scale KSVZ axion (fa = 10 GeV) low). At cryogenic temperatures (T < 60 mK), thermal 3 −23 ∼ with VB = 100 m and Bmax = 5 T is 2.5 × 10 T. To current and voltage noise are subdominant to the cur- detect such a small B-field at this frequency, we need a √ rent shot noise SJ,0 in the SQUID tunnel junctions [43], −19 flux noise sensitivity of 1.2 × 10 Wb/ Hz for a mea- which sets an absolute (temperature-independent) floor surement time of 1 year in a broadband strategy (see for the magnetometer noise. See the appendix for a more below). The anticipated reach for various VB and Bmax detailed discussion of noise in a real implementation of is summarized in Fig. 2. this design. Broadband approach—In an untuned magnetometer, a A typical, temperature-independent flux noise for com- change in flux through the superconducting pickup loop mercial SQUIDs at frequencies greater than ∼10 Hz is induces a supercurrent in the loop. As shown in Fig. 3 √ 1/2 −6 (left), the pickup loop (inductance Lp) is connected in SΦ,0 ∼ 10 Φ0/ Hz, (12) series with an input coil Li, which is inductively coupled −15 to the SQUID (inductance L) with mutual inductance M. where Φ0 = h/(2e) = 2.1×10 Wb is the flux quantum. The flux through the SQUID is proportional to the flux We use this noise level and a fiducial temperature of 0.1 through the pickup loop and is maximized when Li ≈ K as our benchmark. DC SQUIDS are also known to 4 exhibit 1/f noise which dominates below about 50 Hz In a practical implementation√ of an LC circuit with at 0.1 K [47]. We estimate the reach of our broadband resonant frequency ω = 1/ LC, the capacitor has strategy down to 1 Hz assuming 1/f noise is the sole nonzero intrinsic resistance R. Therefore, the circuit has −1 irreducible source of noise at these low frequencies, but in a finite bandwidth ∆ωLC = ω/Q0, where Q0 = (ωCR) . a realistic experiment, environmental noise would likely To maximize the axion signal given the expected band- contribute as well; see the Supplementary Material for width ∆ω/ω ' 10−6, the intrinsic bandwidth of the res- more details. onant circuit should be set to ∆ωLC = max[∆ω, 2π/∆t], Following [25], the signal-to-noise ratio S/N improves where ∆t is the interrogation time at this frequency. 6 with integration time t as While Q0 ' 10 is optimal for sufficiently large ω, smaller Q values are needed at smaller ω to make sure the band- 1/4 1/2 S/N ∼ |ΦSQUID| (tτ) /SΦ,0 (13) width matches the interrogation time. For example, in the strategy of Ref. [24] where each e-fold of frequency 2 for t > τ, where τ is the axion coherence time. The is scanned for a time period te−fold, and thus ∆t = 6 p axion coherence time is approximately te−fold/Q0, one must take Q0 = min[10 , ω te−fold/2π]. Decreasing Q0, however, means adding additional resis- 2π 2π 10−12 eV 6 4 (14) tance to the circuit and thereby increasing the thermal τ ∼ 2 ∼ 10 ≈ 3 × 10 s , mav ma ma noise. Alternatively, we can employ the feedback damping −3 where we have taken v ∼ 10 as the local DM virial circuit of Refs. [48, 49], which can widen the intrinsic velocity. We assume a fiducial integration time of t = bandwidth of the resonant circuit without introducing 1 year, so that t  τ over most of the mass range of additional noise. This allows a large Q factor at all interest. We also assume a geometry with r = R = a = frequencies while still capturing all of the signal [42]. h/3 and a pickup loop inductance L = L . Then, p min The intrinsic Q0 of a niobium superconducting LC cir- requiring S/N > 1 after time t implies sensitivity to 6 6 cuit is over 10 , so we assume Q0 = 10 as our bench- mark, though larger Q may be possible. The signal flux  1/4 0 −18 −1 ma 1 year 5 T through the SQUID depends sensitively on the details of gaγγ > 6.3 × 10 GeV −12 10 eV t Bmax the feedback circuit, but our signal-to-noise analysis will s 1/2 not depend on those details, so we treat the feedback cir- 0.85 m5/2 0.3 GeV/cm3 S × Φ,0√ . cuit as a black box with some inductive coupling M to −6 R ρDM 10 Φ0/ Hz the SQUID in Fig. 3 (right). (15) 8 For Q0 up to ∼10 , thermal noise in the pickup loop dominates over magnetometer noise (see related studies As shown in Fig. 2, an ideal broadband setup with the in Refs. [24, 50] and further discussion in the Supple- benchmark parameters in Eq. (15) could begin to probe mentary Material). Once we know that thermal noise < 14 the QCD axion band for fa ∼ 10 GeV, which is not is dominant, we can calculate the signal-to-noise ratio far below the GUT scale. The sensitivity improves for without regard to the identity of the black box. Fol- larger magnetic fields or larger toroids; for a toroid with 3 lowing Ref. [24], the axion sensitivity is set by requir- VB = 100 m , one can probe the QCD axion band at ing the signal power dissipated in the resonant circuit to the GUT scale. However, larger experiments may not be be greater than that of the noise. The predicted con- sensitive to axion masses near 10−6 eV because displace- straints on gaγγ depend on how much time is spent on ment currents may partially cancel the axion-sourced each frequency band. We imagine a strategy similar to flux. Note that the sensitivity to gaγγ increases at smaller Ref. [24] where each e-fold of frequency is scanned for ma, due to the increase in axion coherence time. a time period te−fold. To compare with the broadband Resonant approach—We now turn to an analysis of a circuit, we take te−fold = 20 days to cover the frequency tuned magnetometer, shown in Fig. 3 (right). This read- −15 range between 1 Hz (ma = 4 × 10 eV) and 100 MHz out circuit has the advantage of enhancing the signal by −7 (ma = 4 × 10 eV) in the same integration time of 1 the quality factor Q at the resonant frequency. The tuned year. circuit is a standard design for detecting small magnetic At frequency m , the signal and noise powers are fields at a given frequency (see e.g. Ref. [43]). Similar a tuned circuits have been considered before for axion DM 2 maΦ r m detection [22] and dark-photon DM detection [24]; our P = Q pickup ,P = k T a , (16) S 0 2L N B 2πt analysis follows closely those of Refs. [24] and [42]. T e−fold

where LT = Lp + Li is the total inductance of the reso- nant circuit. To compare with the broadband reach we √ assume LT = Lmin as in Eq. (11) and take h = 3R. Re- 2 When t < τ, S/N ∼ |Φ | t/S1/2. SQUID Φ,0 quiring a signal-to-noise ratio of unity implies sensitivity 5 to research program DE-SC-0006389, and by a Sloan Re-

 −12 1/4 search Fellowship from the Alfred P. Sloan Foundation. −17 −1 10 eV 20 days gaγγ > 9.0 × 10 GeV ma te−fold s Potential noise sources and experimental details 5 T 0.85 m5/2 0.3 GeV/cm3 106 T × , B R ρ Q 0.1 K max DM 0 In our analysis, we estimated the magnetometer noise (17) as (12) and claimed that it dominated in the broad- where we have assumed a feedback damping circuit that band circuit. This noise level is only a factor of 2 or so above the theoretically predicted temperature- allows us to keep Q0 fixed at low masses. At high masses, the feedback damping circuit is not necessary independent floor from current shot noise [43]. The spec- 6 tral density of shot noise is approximately unless Q0 > 10 is achievable. As illustrated in Fig. 2, the sensitivity increases at 11 Φ S = eI ,I = 0 , (18) larger ma since the signal power density grows as ma. J,0 2 0 0 2L On the other hand, at small masses the broadband ap- proach has a superior projected reach for the same exper- where I0 is the critical current per Josephson junction in imental parameters. Thus, the resonant and broadband an ideal SQUID. This translates into a flux noise of approaches are complementary. r11 √ We introduced a new experimental design that is sen- S1/2 = LS1/2 = hL/ Hz, (19) Φ J,0 8 sitive to ultralight DM with axion-like couplings to elec- −14 −6 tromagnetism in the mass range ma ∈ [10 , 10 ] eV. where h is Planck’s√ constant. Since the signal and shot Most existing axion detection proposals use some kind of noise both scale as L (see (9)), the signal-to-noise ratio resonant enhancement, but we have shown that broad- is largely independent of the SQUID parameters. band circuits can have superior sensitivity for lighter ax- In a real implementation of our experimental design, ion masses. This conclusion agrees with previous lit- magnetic shielding of the entire apparatus will be neces- erature establishing that untuned SQUID magnetome- sary to reduce environmental noise. The thermal motion ters outperform tuned magnetometers at low frequencies of electrons in the shielding material will itself cause ther- [42, 43]; this fact has been exploited in, e.g., Refs. [51, 52] mal noise, however, with an amplitude proportional to to detect fT magnetic fields from MRI experiments with 1/d, where d is the distance from the shield [53]. With a biological tissue samples. A concrete experiment would superconducting shield, this effect is expected to be small likely proceed in two stages: a broadband search over because the only source of thermal noise comes from the a large frequency range, followed by a resonant scan at thin layer of normal material at the outside of the shield. high frequencies and in specific frequency bands if a sig- Moreover, a superconducting shield would significantly nal is seen. We expect that a broadband magnetometer reduce static fluxes compared to a normal conductor such could also be relevant for detecting dark photon DM [24], as copper. With a sufficiently large shield cooled to suffi- and we look forward to further applications of broadband ciently low temperatures, we expect that shielding noise techniques to light DM detection. will be subdominant at frequencies above 1 kHz [54]. As Acknowledgments—We thank Saptarshi Chaudhuri, long as the shield dimensions are on the order of the Kent Irwin, Jeremy Mardon, Lyman Page, Mike Roma- toroid size, the signal flux lines will not be significantly lis, and Chris Tully for detailed discussions of experi- distorted at the center of the toroid, and the signal should mental considerations. In particular, we thank Chris be relatively unaffected. Tully and Mike Romalis for pointing out that we may An additional source of noise may arise from the static reduce the geometric inductance of the pickup loop by current creating the toroidal B-field. In the ideal sce- using smaller loops in parallel, and we thank Kent Ir- nario, this current does not source any magnetic flux win for additionally pointing out that there is a min- through the center of the toroid, which is a benefit com- imal pickup-loop inductance allowed by energy conser- pared to the geometry studied in Ref. [22]. One rea- vation. We thank Asimina Arvanitaki, Dmitry Budker, son this is beneficial is that large fields may make it Simon Coop, Marat Freytsis, Joe Formaggio, Peter Gra- difficult to maintain the pickup loop in a superconduct- ham, Chris Hill, David E. Kaplan, Rafael Lang, Mari- ing phase. However, a non-uniform geometry, combined angela Lisanti, David Pinner, and Surjeet Rajendran for with thermal noise in the toroid, may induce static and helpful conversations. YK thanks Adam Anderson and time-varying flux through the pickup loop due to a small Bill Jones for enlightening discussions regarding SQUIDs. component of the current which circulates azimuthally. BRS is supported by a Pappalardo Fellowship in Physics We expect this source of noise to be subdominant in the at MIT. The work of JT is supported by the U.S. De- kHz-GHz range compared to magnetometer noise (in the partment of Energy (DOE) under cooperative research broadband circuit) or thermal pickup noise (in the res- agreement DE-SC-00012567, by the DOE Early Career onant circuit). One possibility for addressing the fringe 6

fields is to circulate a biasing current in the toroid to In this circuit, the power spectral density of flux noise cancel any static flux though the pickup loop, but this through the SQUID, SΦ(f), at frequency f contains three may itself introduce additional thermal noise. While we contributions [42], namely thermal noise, neglect these noise sources in our analysis, it is impor- tant to carefully consider them in a real implementation T 4kBTLT SΦ (f) = 2 , (20) of the detector. Ns ωQ0 To mitigate the effects of 1/f noise below 50 Hz, one SQUID voltage noise, could attempt to modulate the flux signal either by mod- ulating the toroidal B-field or mechanically modulating L2 (∆ω)2 SV (f) ≈ T S (f), (21) the pickup loop. Such modulation would likely lead to Φ N 2ω2M 2V 2 V additional sources of noise, which must be considered in s Φ a practical design. We do not attempt to evaluate the and SQUID current noise, contribution of 1/f noise in the resonant circuit, which 2 depends on the details of the SQUID coupling, but note J M SΦ(f) ≈ 2 SJ (f). (22) that such a contribution will increase at low tempera- Ns tures below 50 kHz, potentially negating the advantages 2 of operating at lower temperatures. Here, M = α NiL is the coupling of the input inductor Li to the SQUID, ω is the resonant frequency, ∆ω is the bandwidth of the resonant circuit including the contri- −1 Dominance of thermal noise for resonant circuits bution of feedback damping, and Q0 = (ωCR) is the intrinsic quality factor of the capacitor. VΦ characterizes When treating the resonant strategy in the body of the voltage response of the SQUID to a change in flux, 9 10 −1 this letter, we argued that thermal noise in the pickup and it is roughly expected to be R/L ∼ 10 − 10 s . loop dominates over magnetometer noise. Here, we illus- For a SQUID of junction resistance R, SV (f) ≈ 16kBTR trate this observation using the feedback damping circuit and SJ (f) ≈ 11kBT/R + SJ,0 are the spectral densities of Refs. [48, 49], which is one example of the black box of voltage and current noise [43]. We have explicitly in Fig. 3 (right) that couples the LC resonant circuit added the irreducible shot noise contribution SJ,0 to the to the SQUID magnetometer. The effect of the feed- SQUID current noise (see (18)); this term is negligible back circuit is to increase the bandwidth—in our case, at high temperatures but begins to dominate below ∼60 to ∆ω/ω = max[10−6, 2π/(∆t ω)], where ∆t is the in- mK. terrogation time at frequency ω—without increasing the The optimal number of turns Ns is determined by min- noise. We note that this same conclusion, regarding the imizing the magnetic flux noise through the SQUID with dominance of thermal noise, was reached for the case of respect to Ns [42], yielding an inductive shunt circuit with a DC SQUID in [24]. Sim-  α2ωV ilarly, Ref. [24] considered an AC SQUID readout above 2 Φ Ns Lp =Li 1 + 2 4Q0(∆ω) 10 MHz, where the SQUID was biased by a microwave- (23) frequency source in order to maintain a sufficiently large α4Rω2 11k T 1/2 + B + S . Q, and thermal noise was dominant in that case as well. 2 J,0 16kBT (∆ω) R Experimentally, the dominance of thermal noise has been demonstrated for Q up to 106 and T down to 1.2 K for a For ∆ω/ω = 10−6, α2 = 0.5, frequencies ω < 10 MHz, 6 mechanical-electrical resonator designed to detect gravi- and Q0 > 10 , the last term in (23) dominates, giv- 2 6 6 tational waves [50]. ing Ns Lp ≈ 10 Li and thus LT ≈ 10 Li. Depending For the feedback damping circuit, it is useful to gen- on the maximum attainable capacitance, the optimal Ns eralize Fig. 3 to allow the input pickup loop, with in- may be quite large at low frequencies. As an example, ductance Lp, to be inductively coupled to an LC circuit, Ref. [42] estimates a maximum low-loss capacitance of 2 5 whose inductor has inductance Ns Lp. The total induc- 0.1 µF, such that Ns ' 10 for ω = 10 Hz, given Lp as tance of the LC circuit, including the coupling to the calculated in (10). 2 SQUID, is approximately LT ≈ Ns Lp + Li. The separa- Substituting (23) into (20)–(22) gives tion of the pickup loop from the resonant circuit is useful  −6 because, as we will see below and in particular at low 4kBTLp 4 × 10 Q0∆ω ∆ω SΦ(f) ≈ 1 + 2 (24) frequencies, the optimal Ns may be quite large, in order ωQ0 α ω VΦ to minimize thermal noise, while the optimal choice of   −6 2 ω 11 SJ,0 L is always the smallest possible as allowed by energy +10 Q0α + , p VΦ 4 4kBT conservation. Separating the LC circuit from the pickup loop might help mitigate parasitic capacitances. See [24] where the three terms correspond to thermal noise, for more details. SQUID voltage noise, and SQUID current noise. For 7 the parameters of interest, the second term is always ticles,” Ann. Rev. Nucl. Part. Sci. 65, 485–514 (2015), < −2 arXiv:1602.00039 [hep-ex]. subdominant to the third term. Since ω/VΦ ∼ 10 , the third term is suppressed compared to the first for [12] Stephen J. Asztalos et al. (ADMX), “Large scale mi- Q < 108. As discussed in Ref. [24], Q for a niobium crowave cavity search for dark matter axions,” Phys. Rev. 0 ∼ 0 D64, 092003 (2001). superconducting LC circuit is at least 106, but achiev- 8 [13] S. J. Asztalos, G. Carosi, C. Hagmann, D. Kinion, K. van ing Q0 of 10 is difficult. Thus, thermal noise in the LC Bibber, M. Hotz, L. J Rosenberg, G. Rybka, J. Hoskins, resonant circuit dominates the flux noise in the tuned J. Hwang, P. Sikivie, D. B. Tanner, R. Bradley, and magnetometer below 100 MHz, as anticipated. J. Clarke, “SQUID-Based Microwave Cavity Search for It is useful to make contact with the untuned mag- Dark-Matter Axions,” Phys. Rev. Lett. 104, 041301 netometer in this framework. Ignoring for the moment (2010). the finite bandwidth of the signal, as we imagine taking [14] T. M. Shokair et al., “Future Directions in the Mi- crowave Cavity Search for Dark Matter Axions,” Int. Q0 → ∞ at fixed L and C, the resistance in the reso- J. Mod. Phys. A29, 1443004 (2014), arXiv:1405.3685 nant circuit disappears and magnetometer noise should [physics.ins-det]. dominate. Indeed, in that limit the first term in (24) is [15] J. I. Read, “The Local Dark Matter Density,” J. Phys. suppressed, leaving dominantly the current noise, as we G41, 063101 (2014), arXiv:1404.1938 [astro-ph.GA]. found for the broadband circuit: [16] P. Sikivie, “Experimental Tests of the Invisible Axion,” 11th International Symposium on Lepton and Photon In- 2 2 teractions at High Energies Ithaca, New York, August SΦ(f)| = (11kBT + SJ,0)M /Ns . (25) Q0→∞ 4-9, 1983, Phys. Rev. Lett. 51, 1415–1417 (1983), [Erra- Note that this equation refers to flux noise through the tum: Phys. Rev. Lett.52,695(1984)]. SQUID, and (9) can be used to determine the input flux [17] Michael Dine, Willy Fischler, and Mark Srednicki, “A 2 2 2 Simple Solution to the Strong CP Problem with a Harm- noise. Also note that M ∝ Ni ∝ Ns , so that SΦ is less Axion,” Phys. Lett. B104, 199 (1981). independent of Ns in the broadband case. [18] A. R. Zhitnitsky, “On Possible Suppression of the Axion Hadron Interactions. (In Russian),” Sov. J. Nucl. Phys. 31, 260 (1980), [Yad. Fiz.31,497(1980)]. [19] Jihn E. Kim, “Weak Interaction Singlet and Strong CP Invariance,” Phys. Rev. Lett. 43, 103 (1979). ∗ [email protected] [20] Mikhail A. Shifman, A. I. Vainshtein, and Valentin I. † [email protected] Zakharov, “Can Confinement Ensure Natural CP Invari- ‡ [email protected] ance of Strong Interactions?” Nucl. Phys. B166, 493 [1] John Preskill, Mark B. Wise, and Frank Wilczek, “Cos- (1980). mology of the Invisible Axion,” Phys. Lett. B120, 127– [21] K Grohmann, HD Hahlbohm, H L¨ubbig, and H Ramin, 132 (1983). “Current comparators with superconducting shields,” [2] L. F. Abbott and P. Sikivie, “A Cosmological Bound on Cryogenics 14, 499–502 (1974). the Invisible Axion,” Phys. Lett. B120, 133–136 (1983). [22] P. Sikivie, N. Sullivan, and D. B. Tanner, “Proposal [3] Michael Dine and Willy Fischler, “The Not So Harmless for Axion Dark Matter Detection Using an LC Circuit,” Axion,” Phys. Lett. B120, 137–141 (1983). Phys. Rev. Lett. 112, 131301 (2014), arXiv:1310.8545 [4] R. D. Peccei and Helen R. Quinn, “CP Conservation in [hep-ph]. the Presence of Instantons,” Phys. Rev. Lett. 38, 1440– [23] Scott Thomas and Blas Cabrera, “Detecting string-scale 1443 (1977). QCD axion dark matter,” Conference talk at Axions [5] R. D. Peccei and Helen R. Quinn, “Constraints Imposed 2010. by CP Conservation in the Presence of Instantons,” Phys. [24] Saptarshi Chaudhuri, Peter W. Graham, Kent Irwin, Rev. D16, 1791–1797 (1977). Jeremy Mardon, Surjeet Rajendran, and Yue Zhao, “Ra- [6] Steven Weinberg, “A New Light Boson?” Phys.Rev.Lett. dio for hidden-photon dark matter detection,” Phys. Rev. 40, 223–226 (1978). D92, 075012 (2015), arXiv:1411.7382 [hep-ph]. [7] Frank Wilczek, “Problem of Strong P and T Invariance in [25] Dmitry Budker, Peter W. Graham, Micah Ledbetter, the Presence of Instantons,” Phys.Rev.Lett. 40, 279–282 Surjeet Rajendran, and Alex Sushkov, “Proposal for a (1978). Cosmic Axion Spin Precession Experiment (CASPEr),” [8] Peter Svrcek and Edward Witten, “Axions In String The- Phys. Rev. X4, 021030 (2014), arXiv:1306.6089 [hep-ph]. ory,” JHEP 06, 051 (2006), arXiv:hep-th/0605206 [hep- [26] Oliver K. Baker, Michael Betz, Fritz Caspers, Jo- th]. erg Jaeckel, Axel Lindner, Andreas Ringwald, Yannis [9] Rouven Essig et al., “Working Group Report: New Semertzidis, Pierre Sikivie, and Konstantin Zioutas, Light Weakly Coupled Particles,” in Community Summer “Prospects for Searching Axion-like Particle Dark Matter Study 2013: Snowmass on the Mississippi (CSS2013) with Dipole, Toroidal and Wiggler Magnets,” Phys. Rev. Minneapolis, MN, USA, July 29-August 6, 2013 (2013) D85, 035018 (2012), arXiv:1110.2180 [physics.ins-det]. arXiv:1311.0029 [hep-ph]. [27] Peter W. Graham and Surjeet Rajendran, “Axion Dark [10] David J. E. Marsh, “Axion Cosmology,” (2015), Matter Detection with Cold Molecules,” Phys. Rev. D84, arXiv:1510.07633 [astro-ph.CO]. 055013 (2011), arXiv:1101.2691 [hep-ph]. [11] Peter W. Graham, Igor G. Irastorza, Steven K. Lam- [28] Peter W. Graham and Surjeet Rajendran, “New Observ- oreaux, Axel Lindner, and Karl A. van Bibber, “Ex- ables for Direct Detection of Axion Dark Matter,” Phys. perimental Searches for the Axion and Axion-Like Par- Rev. D88, 035023 (2013), arXiv:1306.6088 [hep-ph]. 8

[29] Dieter Horns, Joerg Jaeckel, Axel Lindner, Andrei servatory,” (2013) arXiv:1302.3273 [physics.ins-det]. Lobanov, Javier Redondo, and Andreas Ringwald, [42] Whittier Myers, Daniel Slichter, Michael Hatridge, Sarah “Searching for WISPy Cold Dark Matter with a Dish An- Busch, Michael M¨oßle, Robert McDermott, Andreas tenna,” JCAP 1304, 016 (2013), arXiv:1212.2970 [hep- Trabesinger, and John Clarke, “Calculated signal-to- ph]. noise ratio of MRI detected with SQUIDs and faraday [30] Dieter Horns, Axel Lindner, Andrei Lobanov, and An- detectors in fields from 10µT to 1.5 T,” Journal of Mag- dreas Ringwald, “WISPers from the Dark Side: Ra- netic Resonance 186, 182–192 (2007). dio Probes of Axions and Hidden Photons,” in 9th [43] John Clarke, Claudia D. Tesche, and R.P. Giffard, “Op- Patras Workshop on Axions, WIMPs & WISPs (PA- timization of dc squid voltmeter and magnetometer cir- TRAS13) Mainz, Germany, June 24-28, 2013 (2013) cuits,” Journal of Low Temperature Physics 37, 405–420 arXiv:1309.4170 [physics.ins-det]. (1979). [31] Y. V. Stadnik and V. V. Flambaum, “Axion-induced ef- [44] D Shoenberg, Superconductivity (Cambridge University fects in atoms, molecules, and nuclei: Parity nonconser- Press, 1954). vation, anapole moments, electric dipole moments, and [45] JE Zimmerman, “Sensitivity enhancement of supercon- spin-gravity and spin-axion momentum couplings,” Phys. ducting quantum interference devices through the use of Rev. D89, 043522 (2014), arXiv:1312.6667 [hep-ph]. fractional-turn loops,” Journal of Applied Physics 42, [32] Christian Beck, “Possible resonance effect of axionic dark 4483–4487 (1971). matter in Josephson junctions,” Phys. Rev. Lett. 111, [46] D Drung, R Cantor, M Peters, HJ Scheer, and H Koch, 231801 (2013), arXiv:1309.3790 [hep-ph]. “Low-noise high-speed dc superconducting quantum in- [33] Christian Beck, “Axion mass estimates from resonant terference device magnetometer with simplified feedback Josephson junctions,” Phys. Dark Univ. 7-8, 6–11 (2015), electronics,” Applied Physics Letters 57, 406–408 (1990). arXiv:1403.5676 [hep-ph]. [47] SM Anton, JS Birenbaum, SR O’Kelley, V Bolkhovsky, [34] Jooyoo Hong, Jihn E. Kim, Soonkeon Nam, and Yan- DA Braje, G Fitch, M Neeley, GC Hilton, H-M Cho, nis Semertzidis, “Calculations of resonance enhance- KD Irwin, et al., “Magnetic flux noise in DC SQUIDs: ment factor in axion-search tube-experiments,” (2014), Temperature and geometry dependence,” Physical Re- arXiv:1403.1576 [hep-ph]. view Letters 110, 147002 (2013). [35] B. M. Roberts, Y. V. Stadnik, V. A. Dzuba, V. V. [48] HC Seton, DM Bussell, JMS Hutchison, and DJ Lurie, Flambaum, N. Leefer, and D. Budker, “Limiting P- “Use of a DC SQUID receiver preamplifier in a low field odd interactions of cosmic fields with electrons, protons MRI system,” Applied Superconductivity, IEEE Trans- and neutrons,” Phys. Rev. Lett. 113, 081601 (2014), actions on 5, 3218–3221 (1995). arXiv:1404.2723 [hep-ph]. [49] HC Seton, JMS Hutchison, and DM Bussell, “Gradiome- [36] B. M. Roberts, Y. V. Stadnik, V. A. Dzuba, V. V. Flam- ter pick-up coil design for a low field SQUID-MRI sys- baum, N. Leefer, and D. Budker, “Parity-violating inter- tem,” Magnetic Resonance Materials in Physics, Biology actions of cosmic fields with atoms, molecules, and nu- and Medicine 8, 116–120 (1999). clei: Concepts and calculations for laboratory searches [50] Michele Bonaldi, Paolo Falferi, Massimo Cerdonio, An- and extracting limits,” Phys. Rev. D90, 096005 (2014), drea Vinante, Rita Dolesi, and Stefano Vitale, “Thermal arXiv:1409.2564 [hep-ph]. noise in a high Q cryogenic resonator,” Review of scien- [37] Y. V. Stadnik and V. V. Flambaum, “Searching for dark tific instruments 70, 1851–1856 (1999). matter and variation of fundamental constants with laser [51] Andrei N Matlachov, Petr L Volegov, Michelle A Espy, and maser interferometry,” Phys. Rev. Lett. 114, 161301 John S George, and Robert H Kraus, “SQUID detected (2015), arXiv:1412.7801 [hep-ph]. NMR in microtesla magnetic fields,” Journal of Magnetic [38] Christopher T. Hill, “Axion Induced Oscillating Elec- Resonance 170, 1–7 (2004). tric Dipole Moments,” Phys. Rev. D91, 111702 (2015), [52] Robert McDermott, SeungKyun Lee, Bennie Ten Haken, arXiv:1504.01295 [hep-ph]. Andreas H Trabesinger, Alexander Pines, and John [39] Christopher T. Hill, “Axion Induced Oscillating Elec- Clarke, “Microtesla MRI with a superconducting quan- tric Dipole Moment of the Electron,” (2015), tum interference device,” Proceedings of the National arXiv:1508.04083 [hep-ph]. Academy of Sciences of the United States of America [40] Ben T. McAllister, Stephen R. Parker, and Michael E. 101, 7857–7861 (2004). Tobar, “Axion Dark Matter Coupling to Resonant Pho- [53] S-K Lee and MV Romalis, “Calculation of magnetic field tons via Magnetic Field,” (2015), arXiv:1512.05547 [hep- noise from high-permeability magnetic shields and con- ph]. ducting objects with simple geometry,” Journal of Ap- [41] J. K. Vogel et al., “IAXO - The International Axion Ob- plied Physics 103, 084904 (2008). [54] Mike Romalis, private communication.