Dynamics of simultaneously measured non-commuting

Shay Hacohen-Gourgy†,1,2,∗ Leigh S. Martin†,1,2,3, Emmanuel Flurin1,2, Vinay V. Ramasesh1,2, K. Birgitta Whaley3,4, and Irfan Siddiqi1,2 1Quantum Nanoelectronics Laboratory, Department of , University of California, Berkeley CA 94720, USA. 2Center for Coherent Science, University of California, Berkeley CA 94720, USA. 3Berkeley Center for and Computation, Berkeley, California 94720, USA. and 4Department of Chemistry, University of California, Berkeley, California 94720, USA.

In quantum , cause study how notions of contemporary quantum wavefunction collapse that yields precise out- foundations10–14 arise in such settings. comes, whereas for non-commuting observables such as and Heisenbergs In this work, we implement two high quantum effi- limits the intrinsic preci- ciency readouts of the angular momenta about two dif- sion of a state. Although theoretical work1 ferent axes of an artificial -1/2 system and observe has demonstrated that it should be possible to in real-time the resulting dynamics. Importantly, our perform simultaneous non-commuting measure- measurements are designed to be individually quantum ments and has revealed the limits on measure- non-demolition, which ensures that the back-action arises ment outcomes, only recently2,3 has the dynam- only from the competition between incompatible observ- ics of the been discussed. To re- ables. alize this unexplored regime, we simultaneously apply two continuous quantum non-demolition probes of non-commuting observables to a super- Our experiment utilizes a single superconducting conducting . We implement multiple read- qubit coupled dispersively to a multimode out channels by coupling the qubit to multiple waveguide cavity15. This results in a qubit-state depen- modes of a cavity. To control the dent shift of the cavity mode frequency. By applying a observables, we implement a single quadrature microwave tone to a single cavity mode, one can infer the measurement by driving the qubit and apply- qubit state from the phase of the reflected signal16; this ing cavity sidebands with a relative phase that readout scheme has been used extensively with supercon- sets the . Here, we use this approach ducting for quantum information processing, and to show that the uncertainty principle governs also to perform weak measurements17 and track quantum the dynamics of the wavefunction by enforcing trajectories of a single qubit18. In our configuration, each a lower bound on the measurement-induced dis- cavity mode constitutes a measurement channel which turbance. Consequently, as we transition from extracts the projection of the qubit spin along the σz measuring identical to measuring non-commuting axis of the Bloch sphere. observables, the dynamics make a smooth transi- tion from standard wavefunction collapse to lo- The key ingredient to measuring non-commuting ob- calized persistent diffusion and then to isotropic servables is to re-engineer the underlying Hamiltonian so persistent diffusion. Although the evolution of that each cavity mode couples to a controllable axis of the state differs markedly from that of a con- the qubit. We accomplish this with what we call a single ventional measurement, information about both quadrature measurement (SQM), motivated by the back- non-commuting observables is extracted by keep- action evading techniques recently conceived for optome- ing track of the time ordering of the measure- 19,20 arXiv:1608.06652v3 [quant-ph] 25 Oct 2016 chanical systems . The central idea is to drive Rabi ment record, enabling quantum state tomogra- oscillations Ω /2π = 40 MHz on the qubit so that its phy without alternating measurements. Our R Hamiltonian becomes that of an effective low frequency work creates novel capabilities for quantum con- qubit. As depicted in Fig.1a, we apply a pair of sideband trol, including rapid state purification4, adaptive tones detuned above and below the cavity frequency by measurement5,6, measurement-based state steer- Ω , which the qubit then Raman scatters to the cavity ing and continuous quantum error correction7. R resonance. Separately, the red(blue)-detuned sidebands As physical systems often interact continuously would induce cavity-mediated cooling(heating) of the ef- with their environment via non-commuting de- fective qubit21, as illustrated in Fig. 1b. When they are grees of freedom, our work offers a way8,9 to applied concurrently, their relative phase determines the interference between the up- and down-converted pho- tons, which in turn dictates which qubit observable is ∗[email protected]@berkeley.edu encoded in the resulting signal. The dynamics are gov- erned by a coherent sum of the two effective Hamiltonians 2

(a) State tracking (b) State estimation (a) 1 1

z 0 y 0

JPA JPA 1 -1 -1 Input Out1 Out2 -1 0 0 y -1 0 1 x 1-1 x (c) Cavity mode 1

(b) (c) Cavity mode 2

Qubit t Continuous Non-commuting Tomographic measurement reconstruction Y FIG. 2: Validation of simultaneous non-commuting measurements (a) Reconstruction of two quantum trajec- tories initialized at state (green and blue lines) |−i (black X sphere) and their associated tomographic validation (black crosses). Red and blue arrows indicate axis of each measure- ment. Tomography points are taken every 200 ns, and are each linked to the corresponding trajectory time point with a maroon line. All trajectories ending within ±0.11 around a FIG. 1: Multimode single quadrature measurement given point along the plotted trajectory are used for the cor- (SQM) (a) A transmon qubit in an aluminum cavity. Via responding tomography reconstruction. Error bars are gener- the input port, we drive Rabi oscillations at frequency ΩR ated from statistical uncertainty arising in qubit readout. (b) and apply ±ΩR sideband tones to each of the two lowest Estimation of 16 initially unknown state preparations (black cavity modes. Each output is monitored using a lumped- 31 spheres) using non-commuting measurements. Orange cir- element Josephson parametric amplifier (LJPA) . Coupling cles mark 95% confidence intervals, generated by applying is designed so that ∼90% of each signal is routed to its corre- maximum likelihood estimation to ∼10,000 trajectories each. sponding LJPA. This system yields two separate measurement (c) Pulse sequence, showing Rabi drive (green) and sideband channels, each with a controllable measurement . (b) tones applied to cavity modes 1 and 2 (blue and red respec- Dressed state picture of the SQM scheme for one of the cav- tively), which generate measurements indicated by arrows in ity modes. Sideband tones drive transitions indicated by the panels a and b. Tomographic reconstruction consists of a to- solid lines, and undulating lines represent cavity decay, which mography qubit pulse and projective readout. we detect. (c) Tomographic reconstruction after a 1 µs SQM applied only to the lower mode, showing collapse along three separately chosen measurement axes σδ1 , δ1 = {0, π/4, π/2}. Circles plot tomographic data and lines give theory based on axis σδ ≡ σx cos δ + σy sin δ. The resulting Hamiltonian a dephasing rate of Γ1/2π = 122 kHz and quantum efficiency is a resonant cavity drive, the sign of which depends on of η1 = 0.41. the qubit state along the σδ axis. Detecting the cavity output field with quantum efficiency η yields a measure- 2 2 ment of the qubit at a rate Γη = 2χ a¯0η/κ in the σδ basis16 (see methods), which we can now control. First, to demonstrate control of the measurement axis, χa¯0 h † iδ † −iδ −iδ † † iδ i Heff = aσ e + a σe + aσe + a σ e we perform a single SQM using the lower cavity mode. 2 | {z } | {z } resonant heating resonant cooling We prepare the |−i state and read out for 1 µs followed by a tomography rotation and a projective measurement us- χa¯0 = (a + a†)σ (1) ing a standard dispersive readout of σ in the lab frame. 2 δ z During the SQM, drift in the Rabi frequency and spuri- where χ is the qubit-cavity coupling,a ¯0 is the sideband ous tones at the cavity frequency induce unwanted rota- † amplitude, a, a are the cavity ladder operators and σ, tions about the σz and σδ axes respectively (see meth- † σ are the raising and lowering operators√ with respect ods). To mitigate these effects, we actively feedback to to the dressed basis |±i = (|gi ± |ei)/ 2. Note that stabilize the Rabi frequency to within 10 kHz and sup- this scheme implements longitudinal coupling, an impor- press leakage of the local oscillator used to generate the tant feature for squeezing-enhanced readout22. Impor- sidebands to the 10−4 level. Because a single tantly, the relative sideband phase δ sets the coupling SQM commutes with itself at all times, the of 3

(b) 32 ns 256 ns 480 ns 704 ns 928 ns 1600 ns (a) Steady-state 5x10-4

0o 18o 36o Exp 1

y 54o 72o 90o Thy -1 0 32 ns 544 ns 1056 ns 1568 ns 2080 ns 3616 ns -1 x 1

(c) π No

0.4 r t=160ns σ2 mali

Exp z 0 ed c 1 t=288ns 0 1 2 Time (μs) 0.2 t=352ns ou t=224ns y t=480ns n t=672ns ts Thy t=1056ns t=2848ns -1 0 -1 x 1 0 3x10-4 -π angle0 π

FIG. 3: of the (a) in the steady-state as function of angle, between the axes, demonstrating loss of collapse for non-commuting observables. (b) as function of time for perpendicular measurement axes. We prepare a mixed state for observing the radial dynamics (top), and a state with purity of P = 0.89 for observing the azimuthal dynamics (bottom). Upper row of each is experimental data, and bottom row is theoretical comparison derived from the Fokker-Planck equation (see methods). (c) Angular probability distributions for states within a ring of inner radius 0.86 and outer radius 0.92, showing that dynamics match those of a random walk. Points are normalized counts and lines are fits to normal distributions convolved with a 2π periodic Dirac comb. Inset – as a function of time (violet) and linear fit (yellow) yielding a slope of 1.4 µs−1. The expected slope for a perfect random walk in our system is 1.5 µs−1 with the main source of uncertainty due to measurement rate uncertainty of about 10%.

δ - δ = 0 δ - δ = 45o δ - δ = 90o 2 1 2 1 2 1 the signal fully specifies the amount of projection onto 1 1 1 an eigenstate18. We perform tomographic state valida- tion on subsets of the data post-selected according to y0 y0 y0 this integrated signal. Tomography data in Fig. 1c show

1 1 1 state collapse along the desired axis, confirming control of the measurement σ . -1-1 0 z -1-1 0 z -1-1 0 z δ 0 0 0 x 1-1 x 1-1 x 1-1 Next, to simultaneously measure two non-commuting 1 1 1 observables, we apply the SQM to the two lowest cav- ity modes. Each pair of sidebands sets a measurement y 0 y 0 y 0 axis, and thus we are able to control them independently. Because non-commutativity leads to dependence on the time ordering of measurement outcomes, the integral of -1 -1 -1 -1 0 1 -1 0 1 -1 0 1 x x x the output signal is no longer sufficient to uniquely de- termine the qubit state in general. Nevertheless, the 0 0.11 0.22 combined measurement records encode complete infor- mation, and we are able to track quantum trajectories of FIG. 4: Magnitude of the quantum back-action. The amplitude of disturbance induced by measurement an initially known state. We calculate the trajectories by (Tr[dρ†dρ]) at each point on the Bloch sphere for three rel- iteratively applying a measurement operator which corre- ative measurement angles δ2 − δ1. (top) Disturbance at the sponds to the record of both σδ1 and σδ2 integrated over a surface of the Bloch sphere, which is bounded from below short interval ∆t = 16 ns (see methods). To experimen- by the uncertainty principle. (bottom) A slice through the tally verify the trajectories, we perform quantum state z = 0 plane. Dotted lines mark measurement axes. For non- tomography23 on subsets of trajectories that end at ap- commuting observables (δ2 − δ1 6= 0), no point of zero dis- proximately the same point on the Bloch sphere. Figure turbance exists within the entire volume of the Bloch sphere, 2a shows two such traces with their tomographic valida- indicating that the state must diffuse indefinitely. tion from data in which the measurement axes are set to

be orthogonal, σδ1 = σx and σδ2 = σy. 4

Due to the disturbances introduced by measurement patible, implying that the state diffuses indefinitely. If incompatibility, extraction of an initially unknown quan- we calculate this disturbance directly from the stochas- tity, such as a Hamiltonian parameter or system observ- tic master equation for the system, we find the following able, requires use of the combined measurement records relation valid for all initially pure states and their full time orderings. In particular, estimation methods must rely not only on the statistics of the mea- Tr[dρ†dρ] = (∆σ2 Γ η + ∆σ2 Γ η )dt (2) surement records, but also on some estimate of this dis- δ,1 1 1 δ,2 2 2 p turbance. We encode this information in a set of com- ≥ |h[σδ1 , σδ2 ]i| η1η2Γ1Γ2dt posite measurement operators, one for each trajectory24 which holds for any Hermitian operators of any sys- (see methods), and use them to reconstruct a set of ini- tem. The right hand side of the equality closely re- tial state preparations, analogous to state tomography. sembles the original Heisenberg uncertainty relation, but We then perform maximum likelihood estimation of the contains the sum of the instead of the prod- initial state over a large set of trajectories taken with the uct. Unlike the latter, the sum can be bounded by same initial state24. Figure 2b shows reconstruction of a stronger which is never trivial for non- sixteen state preparations again in the case of orthogo- 25 nal measurement axes. Agreement within the confidence commuting measurements . This shows that the per- interval demonstrates that despite the complicated dy- sistent diffusion observed in Fig. 3c is a universal conse- namics induced by competing observables, our scheme quence of the uncertainty principle and can be quantita- performs as a measurement, and that it extracts infor- tively derived from it. mation about σ and σ simultaneously. The techniques we developed readily generalize to x y multi-level and multi-qubit systems, and have numer- A state undergoing a non-commuting measurement ex- ous potential applications for quantum information sci- hibits dynamics beyond those of usual wavefunction col- ence. The ability to measure multiple distinct opera- lapse. We directly observe this evolution by measuring tors of a quantum system simultaneously could allow the probability distributions of the resulting density ma- for implementation of continuous quantum error correc- trices. Figure 3a shows the steady state probability dis- tion, in which several error syndromes can be moni- tributions. When the axes align, the system collapses to tored concurrently7. Furthermore it may motivate de- one of two points at the poles of the Bloch sphere as ex- velopment of quantum-limited metrology protocols that pected for commuting measurement operators. When the acquire information simultaneously about multiple in- axes are separated by a finite angle less than 90 degrees, compatible properties. By monitoring a third cavity the state does not collapse to any point, but rather local- mode and applying an additional sideband drive, the axis izes to regions of finite defined by the measurement perpendicular to the Rabi plane can also be probed26. axes. A salient feature is that when the axes are orthog- With a modest improvement in quantum efficiency, our onal and hence maximally incompatible, the location of scheme will also lead to faster initialization of quantum the measurement axes no longer leaves any imprint on circuits4,27. the state evolution. The distinct regions merge and we Our experiment reveals the subtle interplay between lose all notion of collapse. the Heisenberg uncertainty principle and wavefunction Figure 3b shows probability distributions as a function collapse. As incompatibility of observables is at the core of time for this canonical case. Starting in a mixed state, of quantum theory28, tests of must we see that the state purifies isotropically to a mean √ access these properties. Existing work on quantum foun- steady-state radius given by r = η. High quantum effi- dations has focused on testing the validity of essential ciency in our system allows us to observe the azimuthal features of , such as contextuality dynamics by preparing a state of purity P = 0.89, the and the various bounds on error and disturbance. Our most likely steady-state purity. As predicted by Ruskov work presents the possibility of exploring how such et al.3 in the case of unit quantum efficiency, competition concepts emerge in realistic systems8,9, which tend to between the maximally incompatible observables leads to interact continuously with their environments via mul- diffusive given by a uniform random walk. Figure tiple non-commuting decoherence channels. Thus our 3c shows the angular distribution as a function of time, work provides an indispensable tool for investigating this which agrees quantitatively with this prediction. virtually unexplored territory of quantum foundations. The diffusive behavior seen in Fig. 3c suggests that even once the probability distributions have reached † These authors contributed equally to this work. steady state, the system continues to evolve. To quantify Acknowledgements We thank Justin Dressel, An- this measurement-induced disturbance, we plot the norm drew Jordan, Alexander Korotkov, Joshua Combes, Mo- of the state change dρ over an interval of 64 ns versus po- han Sarovar, Uri Vool and Juan Atalaya for invalu- sition on the Bloch sphere in Fig. 4. When the measure- able discussions, and MIT Lincoln labs for fabrication ments are compatible, two points on the Bloch sphere of the TWPA. L.S.M. and V.V.R. acknowledge exhibit zero disturbance, which indicate mutual eigen- from the National Science Foundation Graduate Fellow- states of the measurements and hence points of collapse. ship Grant No. 1106400. L.S.M. additionally acknowl- No such points exist when the measurements are incom- edges support from the Berkeley Fellowship for Graduate 5

Study. This work was supported by the Air Force Office of Scientific Research under Grant # FA9550-12-1- 0378 x and the Army Research Office under Grant # W911NF- y 15-1-0496. 4 Author Contributions S.H. and L.S.M. wrote the manuscript, conducted the experiment and analyzed the 3 data. S.H., L.S.M., E.F. and I.S. conceived of the ex- y periment. S.H. and L.S.M. constructed the experimen- 2 ap h tal setup with assistance from V.V.R. and E.F.. S.H. r fabricated the qubit and LJPAs. L.S.M. performed theoretical analysis with assistance from S.H. All au- omo g 1 T thors contributed to discussions and preparation of the manuscript. All work was carried out under the supervi- 0 sion of K.B.W. and I.S.. Author Information The authors declare no com- -1 peting financial interests. -1 -0.5 0 0.5 1 SQM

Extended Data Fig 1: SQM vs tomographic validation Tomographic validation of a set of trajectories initialized in the state y = −1 and tracked for 3µs, with the axes being per- pendicular. Horizontal axis indicates Bloch sphere coordinate as predicted by trajectory reconstruction. Vertical axis repre- sents the coordinate reconstructed from post-selected tomog- raphy data. Error bars are derived from the Poison statistics of the qubit measurement. Data are staggered vertically to I. METHODS allow for clearer visualization by adding one of seven arbitrary offsets. These offsets are chosen according to the number of Experimental setup and sample. Our transmon measurements contributing to each data point. qubit is fabricated on double-side-polished silicon, with a single double-angle-evaporated Al/AlOx/Al Josephson junction. The internal dimensions of the aluminum 3D also lost through the input pin. We estimate the total re- cavity are 81 mm x 51 mm x 20 mm. The qubit is posi- duction in quantum efficiency due to these loss channels tioned 23 mm from the edge of the cavity. The qubit is to be approximately 10% for each mode. The total quan- characterized by a charging Ec/h = 220 MHz and tum efficiencies are η1 = 0.41 and η2 = 0.49 for modes 1 a |0i to |1i transition frequency ωq/2π = 4.262 GHz. The and 2 respectively. qubit is characterized by an life- Measurement Calibration. We generate SQM side- time of T1 = 60 µs, echo time of T2,echo = 40 µs and Rabi band tones by mixing a common local oscillator (LO) set decay time of 25 µs. The two lowest cavity modes used to the corresponding cavity frequency with 40 MHz tones in the experiment have frequencies ω1/2π = 6.666 GHz, and controllable DC offsets input to the I and Q ports of ω2/2π = 7.391 GHz, linewidths κ1/2π = 7.2 MHz, κ2/2π a mixer. The output is followed by an amplifier model = 4.3 MHz and qubit dispersive frequency shift χ1/2π = ZRON8G+ and then passed through a home built notch 0.18 MHz, χ2/2π = 0.23 MHz respectively. The cavity filter which attenuates the carrier by ∼ 25 dB relative outputs are amplified using two lumped-element Joseph- to the sideband tones. The filter and DC offset controls son parametric amplifiers (LJPA) operated in phase sen- allow suppression of the LO to the 10−4 photon num- sitive mode. Amplifier gains are set to 15 dB for mode 1 ber level, which eliminates spurious rotations about the and 18 dB for mode 2. To prevent saturation of the am- measurement axis. We balance the relative amplitudes plifiers by the SQM sideband tones, we designed them to of the sideband tones by measuring the induced cooling have relatively narrow bandwidths of 20 MHz bandwidth or heating with Ramsey interferometry21, and then ad- for mode 1 with 7.3 MHz bandwidth for mode 2. The justing the relative phase between the I and Q 40 MHz signals are further amplified with two cryogenic HEMT signals. amplifiers, model number LNF4 8. Due to a high noise Since the sideband tones stark shift the qubit, all qubit of HEMT 1, we added a Josephson traveling drives and pulses are applied with the sidebands present parametric amplifier29 immediately before it. and at the Stark-shifted qubit frequency, so that the ex- In order to route signal from mode i only to the cor- periment takes place entirely in this frame. The Rabi fre- responding LJPA, the pin coupling to mode 1 is placed quency is chosen such that the measurement operates in at the node of mode 2 (see Fig. 1), so that from the unresolved sideband regime ΩR  κ. The sideband mode 2 do not couple to it. Placement of pin 2 and the tone powers are set such that the measurement rate is choice of κ1 > κ2 reduces leakage of mode 1 signal into much slower than the cavity mode bandwidths, so that pin 2. A small amount of signal from modes 1 and 2 is the polaron transformation16 holds when both measure- 6 ments are on. Temperature drifts in the room-temperature - 40 ics lead to instability of the Rabi frequency, which in- duces dephasing along the σz axis of the qubit in the 30 measurement frame. To efficiently and precisely measure this drift, we perform a 1 µs SQM followed by a lab- 20 frame readout of σz. We interleave these measurements between the taking of trajectory data. This duration is 10 Rabi amplitude index chosen so that hσzi ≈ 0, which means that the readout 0 π /2 π 3 π /2 2 π is maximally sensitive to drift in the Rabi frequency. We SQM phase (radians) use this signal to stabilize the Rabi frequency by opti- mizing the power of the Rabi drive every few seconds. Extended Data Fig 2: SQM calibration. Integrated After setting up the SQM sidebands, one must care- SQM signal for measurement calibration. Color axis is scaled fully set the feedback so that it stabilizes the Rabi fre- to the outcomes in which the qubit is aligned or anti-aligned to quency to 40 MHz. Furthermore, as the Rabi drive ramps the measurement axis. Vertical sweep is performed by sweep- up to its maximal power over a finite duration, the rela- ing the modulation amplitude to drive approximately 40 MHz tive angle that the initial qubit state makes with the SQM Rabi oscillations. The red dashed line indicates the index at axis is also unknown and must be measured. We jointly which the Rabi frequency was determined to be 40 MHz. calibrate these parameters by sweeping both the SQM axis and the Rabi frequency. We perform this calibra- of the Gaussians and τ is the mea- tion by measuring the SQM signal integrated over 7 µs surement duration. as a function of these two sweep parameters, as shown In order to validate our system calibra- in extended data Fig. 2. When the initial qubit state tion and quantum trajectory reconstructions, is aligned or anti-aligned with the measurement axis and we apply one of seven tomography pulses the Rabi frequency is precisely 40.00 MHz, then the qubit {Identity, π/2 , −π/2 , π/2 , −π/2 , π , −π } at the remains aligned throughout the duration of the measure- x x y y x x end of each trajectory followed by a dispersive readout in ment, and the integrated output signal takes its maximal the lab frame. Extended data Fig. 1 shows comparison or minimal value. Thus the Rabi frequency is 40.00 MHz between tomographic estimates of hσ i and hσ i and when this contrast is maximal. We have indicated this x y their corresponding estimates from trajectory recon- region of the plot with a red line. Rotation of the qubit struction. To generate the tomographic validation data, during the measurement leads to the slight tilt evident we have partitioned the Bloch sphere into a 15 × 15 × 15 in extended data Fig. 2, which means that a 1D sweep grid for post-selection. Each data point is a tomographic of just the phase or Rabi amplitude would be insuffi- measurement of all trajectories which were predicted to cient to accurately find the global maximum. Due to end within the corresponding voxel. 57% of data points instability of the Rabi frequency without feedback, the lie within error bars, indicating good agreement and low units in the vertical axis of the plot are not precisely contribution of systematic errors. known. While taking these data, we also perform the Single Quadrature Measurement. Our system is measurements described in the previous paragraph for described by the following dispersive Hamiltonian feedback stabilization of the Rabi frequency and record the measurement outcome associated with each setting X † of the Rabi frequency. Once the desired Rabi frequency H = Hq + Hdrive + (ωc,i + χiσz)ai ai (4) is identified, we look up the associated feedback measure- i={1,2} ment outcome and use this for value for stabilization. ω H = q σ + Ω (σ + σ†) cos(ω t) In order to measure quantum trajectories, we require q 2 z R d an accurate estimate of the SQM rates Γi and quantum X † ∗ Hdrive = i(t)ai + i(t) ai efficiencies ηi. The former we measure by preparing |+i and then making a Ramsey measurement. We measure i={1,2} the quantum efficiency by preparing states aligned and where i(t) ∈ C is the coherent drive that will represent anti-aligned with the SQM. Histograms of the integrated the SQM sideband drives. The master equation for a measurement records yield a pair of Gaussians which sep- cavity which is damped at rate κi and monitored with arate as a function of time. The quantum efficiency is quantum efficiency ηi is given by30. √ X iφi dρ = −i[H, ρ]dt + κiD[ai]ρ dt + κiηiH[aie ]ρ dWi 2 (µ↑ − µ↓) i={1,2} ηi = 2 (3) 8τσ Γi (5)

† † † where µ↑/↓ is the mean of the Gaussian for the where D[X]ρ = XρX −(X Xρ+ρX X)/2 is the dissi- aligned/anti-aligned state preparation, σ is the average pation superoperator, H[X]ρ = Xρ+ρX† −hXρ+ρX†iρ 7 and φi is the amplification axis of the phase sensitive am- plifier used to monitor mode i. dWi is a Gaussian dis- tributed variable with variance dt extracted from mea- χa¯0 H = (eiΩRt+iδ + e−iΩRt−iδ)× (9) surement record i. For the derivation, it suffices to con- q-frame 2 sider just one of the two cavity modes. To measure a (d† + d)(σe−iΩRt + σ†eiΩRt) single quadrature of the qubit, we drive the cavity with a h cos(2Ω t + 2δ) i R † −iΩt † iΩRt pair of sidebands detuned by ±ΩR from the cavity reso- + χ n¯0 + d d (σe + σ e ). nance, and with phases ±(δ−atan(κ/2ΩR)). As ΩR  κ, 2 the latter term can be ignored in our experiment. We p 2 2 choose this drive (t) = −ia¯0 ΩR + κ /4 sin(ΩRt + δ − Dropping terms which rotate at ΩR or 2ΩR, we are left −iωct atan(κ/2ΩR))e so that the cavity internal displace- with ment due to the sidebands takes the form

χa¯0 † iδ † −iδ † −iωct Hq-frame = (d + d)(σe + σ e ) =gσ ˜ δ(d + d) a¯(t) =a ¯0 cos(ΩRt + δ)e → a¯0 cos(ΩRt + δ). (6) 2 σδ ≡ cos(δ)σx + sin(δ)σy The right arrow indicates that we have transformed χa¯ g˜ ≡ 0 (10) into the with respect to the cavity. We 2 takea ¯0 to be real for simplicity and define the displaced cavity operator d = a − a¯. We first transform the cavity dissipation terms which is a qubit-state-dependent cavity drive Hamilto- nian. If in addition to the sideband tones, a small amount of LO leakage (photons at the frequency of the mode) κD[a]ρ = κD[d +a ¯]ρ = κD[d]ρ − i[H , ρ] (7) dis is also present at the cavity input, then Eq. 6 would iκ H = − (¯ad† − a¯∗d) readα ¯0 cos(ΩRt+δ)+¯aLO, and the effective Hamiltonian dis 2 √ √ would instead be κηH[aeiφ] = κηH[deiφ].

This change of variables is equivalent to transforming † Hq-frame =gσ ˜ δ(d + d + 2Re[¯aLO]). (11) the Hamiltonian by the displacement operator exp(¯aa† − aa¯ ), and thus adds a term H0 = i(ad¯˙ −ad¯˙ †) to the Hamil- 0 tonian. Note that Hdis and H cancel the drive term Thus LO leakage induces unwanted rotations about the Hdrive from Eq. (4). Writing the remaining terms of H for σδ axis. one cavity mode in the interaction picture with respect to If the cavity starts in the when this the cavity and in the frame of the qubit drive (i.e. trans- Hamiltonian is turned on, then it remains in a coherent † forming the Hamiltonian by exp[iωca at+iωdσzt/2]) and state at all times. To derive an effective master equation applying the rotating wave approximation, we find for the qubit, we must eliminate the cavity entirely from Eq. (5). This may be accomplished by first applying a transformation which displaces the cavity to its ground † 16 Hint = χa aσz + Hq,int (8) state, which is known as the polaron transformation † ∗ = χ[(d +a ¯ )(d +a ¯)]σz + Hq,int

h † n¯0 = χ a¯0 cos(ΩRt + δ)(d + d) + + 2 U = |eδiheδ|D[αe] + |gδihgδ|D[αg] (12) κ cos(2ΩRt + 2δ) † i n¯ + d d σ + H α˙ e/g = −ie/g − αe/g e/g = ±g˜ 0 2 z q,int 2 1 Hq,int = [(ωq − ωd)σz + ΩRσx] 2 where D is the cavity displacement operator, |eδ/gδi are the eigenstates of σδ and αe/g is the cavity displace- 2 where we have definedn ¯0 =a ¯0. Choosing the Rabi ment conditional on the qubit state. Because the second drive to be resonant (i.e. ωq − ωd = −χn¯0), we diago- line is the classical equation of motion for a resonantly nalize the qubit drive term of the Hamiltonian by going driven cavity with drive rate  and damping κ, this trans- into the Hadamard frame (σz ↔ σx), and then going into formation maps the cavity to its ground state at all times, a frame rotating at the Rabi frequency (exp[iΩRσzt/2]). allowing us to trace it out. Transforming back from the These transformations eliminate Hq,int and map σz to polaron frame to the qubit frame, we find that the effec- σe−iΩRt + σ†eiΩRt, so the Hamiltonian becomes tive stochastic master equation for the qubit is 8

√ p 0 ΓM ΓM r dρ = Lρ dt + H[σδ]ρ dW − i [σδ, ρ]dW X Γi Γiηi 2 2 dρ = D[σ ]ρ dt + H[σ ]ρ dW (15) 2 δi 2 δi (13) i={1,2} B Γ dW Lρ = −i [σδ, ρ] + D[σδ]ρ √ i 2 2 Vidt = hσδi idt + 2ηiΓi B = 2˜gRe(αe + αg) Γ = −2˜gIm(αe − αg) q p √ 0 √ ΓM = κη|β| cos(φ − θβ) ΓM = κη|β| sin(φ − θβ) where Vi is the measurement signal at time t, normal- ized appropriately. Equation 15 can also be converted β = αe − αg to a Fokker Planck equation, which propagates proba-

where θβ = arg(β) is the angle of the cavity displace- bility distributions of ρ. We use the latter to generate ment axis β in the IQ plane and Γ is the dephasing theory plots for Fig. 3. This stochastic master equation rate. Substituting the equations of motion for the cavity is generated by the following measurement operator (Eq.12) into the remaining expressions of Eq.13, we find

2   4˜g −κt/2 X Γiηi 2 dρ = (1 − e )D[σδ]ρ dt + (14) Ω(V ) = exp − (V (t) − σ ) dt (16) κ  2 i δ,i  r i=1,2 4˜g2 η(1 − e−κt/2)H[eiφσ ]ρ dW † κ δ Ωρ(t)Ω ρ(t + dt) = E1−ηi † 8˜g2 2χ2n¯ Tr[Ωρ(t)Ω ] =⇒ Γ = = 0 . κ κ We align the LJPA amplification axis to the axis of displacement arising in Eq. 10, so that φ = θβ. Applica- where E1−ηi is a superoperator which models dephas- tion of sidebands to another mode of the cavity does not ing due to finite quantum efficiency. To assure positivity change the above derivation except to add an additional of the state when dt is taken to be finite, we use Eq. 16 measurement also modeled by Eq. 14. Neglecting time- to numerically propagate quantum trajectories and also scales of order 1/κ, measurement of two observables σδ1 to calculate probabilities for maximum likelihood recon- and σδ2 is modeled by struction.

[1] Arthurs, E. & Kelly, J. L. On the Simultaneous Measure- qubit observables. arXiv preprint arXiv:1606.07934 ment of a Pair of Conjugate Observables. Bell System (2016). Technical Journal 44, 725–729 (1965). [9] Nishizawa, A. & Chen, Y. Universally valid error- [2] Jordan, A. N. & B¨uttiker, M. Continuous quantum mea- disturbance relations in continuous measurements. arXiv surement with independent detector cross correlations. preprint arXiv:1506.00304 (2015). Phys. Rev. Lett. 95, 220401 (2005). [10] Kirchmair, G. et al. State-independent experimental test [3] Ruskov, R., Korotkov, A. N. & Mølmer, K. Qubit state of quantum contextuality. Nature 460, 494–497 (2009). monitoring by measurement of three complementary ob- [11] Kochen, S. & Specker, E. P. The Problem of Hidden servables. Phys. Rev. Lett. 105, 1–4 (2010). Variables in Quantum Mechanics. J Math. Mech. 17(1), [4] Ruskov, R., Combes, J., Mølmer, K. & Wiseman, H. M. 59–87 (1968). Qubit purification speed-up for three complementary [12] Erhart, J. et al. Experimental demonstration of a univer- continuous measurements. Philosophical Transactions of sally valid error-disturbance uncertainty relation in spin the Royal Society A: Mathematical, Physical and Engi- measurements. Nat. Phys. 8, 185–189 (2012). neering Sciences 370, 5291–5307 (2012). [13] Dressel, J. & Nori, F. Certainty in Heisenberg’s uncer- [5] Jacobs, K. How to project qubits faster using quantum tainty principle: Revisiting definitions for estimation er- feedback. Phys. Rev. A 67, 030301 (2003). rors and disturbance. Phys. Rev. A 89, 022106 (2014). [6] Wiseman, H. M. Adaptive phase measurements of optical [14] Ozawa, M. Universally valid reformulation of the Heisen- modes: Going beyond the marginal q distribution. Phys. berg uncertainty principle on noise and disturbance in Rev. Lett. 75, 4587 (1995). measurement. Phys. Rev. A 67, 042105 (2003). [7] Ahn, C., Doherty, A. C. & Landahl, A. J. Continuous [15] Koch, J. et al. Charge-insensitive qubit design derived via quantum feedback control. from the Cooper pair box. Phys. Rev. A 76, 042319 Phys. Rev. A 65, 042301 (2002). (2007). [8] Luis Pedro Garca-Pintos, J. D. Probing quantumness [16] Gambetta, J. et al. Quantum trajectory approach to with joint continuous measurements of non-commuting circuit QED: Quantum jumps and the Zeno effect. Phys. 9

Rev. A 77, 012112 (2008). taneous measurements, imperfections, and decoherence. [17] Hatridge, M. et al. Quantum back-action of an individ- Phys. Rev. A 93, 012109 (2016). ual variable-strength measurement. Science 339, 178–81 [25] MacCone, L. & Pati, A. K. Stronger uncertainty relations (2013). for all incompatible observables. Phys. Rev. Lett. 113, 1– [18] Murch, K. W., Weber, S. J., Macklin, C. & Siddiqi, I. Ob- 5 (2014). arXiv:1407.0338. serving single quantum trajectories of a superconducting [26] Vool, U. et al. Continuous quantum nondemolition quantum bit. Nature 502, 211–214 (2013). measurement of the transverse component of a qubit. [19] Caves, C. M., Thorne, K. S., Drever, R. W., Sandberg, arXiv:1605.08004 (2011). arXiv:1605.08004. V. D. & Zimmermann, M. On the measurement of a weak [27] Combes, J., Wiseman, H. M. & Scott, A. J. Replacing classical force coupled to a quantum-mechanical oscilla- quantum feedback with open-loop control and quantum tor. Rev. Mod. Phys. 52, 341 (1980). filtering. Phys. Rev. A 81, 020301 (2010). [20] Hertzberg, J. B. et al. Back-action-evading measure- [28] Dressel, J. Indirect Observable Measurement: an Alge- ments of nanomechanical motion. Nat. Phys. 6, 213–217 braic Approach, PhD Thesis, Univ of Rochester (P52) (2010). (2013). [21] Murch, K. W. et al. Cavity-assisted quantum bath [29] Macklin, C. et al. A near quantum-limited Josephson engineering. Phys. Rev. Lett. 109, 1–5 (2012). traveling-wave parametric amplifier. Science 350, 307 arXiv:1207.0053. (2015). [22] Didier, N., Bourassa, J. & Blais, A. Fast quantum non- [30] Korotkov, A. N. Quantum Bayesian approach to cir- demolition readout by parametric modulation of longitu- cuit QED measurement. arXiv:1111.4016 11 (2011). dinal qubit-oscillator interaction. Phys. Rev. Lett. 115, arXiv:1111.4016. 203601 (2015). [31] Castellanos-Beltran, M. A., Irwin, K. D., Hilton, G. C., [23] Nielson, M. A. & Chuang, I. L. Quantum computation Vale, L. R. & Lehnert, K. W. Amplification and squeez- and quantum information (Cambridge University Press, ing of with a tunable Josephson metama- 2010). terial. Nat. Phys. 4, 929–931 (2008). [24] Six, P. et al. Quantum state tomography with noninstan-