Experimental state and process reconstruction. Philipp Schindler + Thomas Monz Institute of Experimental Physics University of Innsbruck, Austria

Transcription

1 Experimental state and process reconstruction Philipp Schindler + Thomas Monz Institute of Experimental Physics University of Innsbruck, Austria

2 What do we want to do? Debug and characterize a quantum computer Prove properties of the generated state Prove properties of the implemented process

3 Outline Tomographic reconstruction Our quantum information processor Entanglement assisted tomography State tomography with statistical noise Noise + errors in our setup Detection of systematic errors in tomographic data

4 Tomos (greek) = part, section Graphein (greek) = to write Look at section/projections, Use reconstruction method to infer the object of desire Picture from Wiki

5 States and Processes Describe state via density matrix ρ Describe quantum process via χ matrix

6 States and Processes Describe state via density matrix ρ Describe quantum process via χ matrix Dimensions: 2N x 2N Dimensions: 4N x 4N Properties: Pure? Entangled? Properties: Unitary? Local / Entangling? >> Hilbert space is a big space <<, Carlton Caves

7 Choi-Jamiolkowski isomorphism Every process can uniquely be mapped to a higher-dimensional density matrix All distance measures for states are applicable to quantum processes. (In principle) All state tomography techniques can be applied to infer quantum processes Reports on Mathematical Physics 3, 275 (1972) Linear Algebra and its Applications 10, 285 (1975)

8 Distance measures Metric: should be a metric (E,F) 0 with (E,F) =0 if and only if E=F (E,F) = (F,E) (E,G) (E,F) + (F,G) Easy to calculate Easy to measure Physical interpretation Phys. Rev. A 71, (2005)

9 (Uhlmann) Fidelity Fulfills all requirements: Can be calculated Can directly be measured (with respect to pure states) Interpretation: Probability to find the desired state in the system Fidelity does not fulfill F(ρ,ρ) = 1 for all possible ρ the 'standard' fidelity definition does not return F(ρ,ρ)=1 for mixed states replace 'standard fidelity' by Uhlmann fidelity, that also works for mixed states Fidelity is still not a metric Rep. Math. Phys. 9, 273 (1976)

10 Trace Distance Trace distance is a metric Interpretation differs significantly: Worst-case scenario Given ρ and σ, and the best suited observable to distinguish the states What is the probability of being able to distinguish the states? Often used by theorists, less by experimentalists % fidelity sounds better than ~ 30 % probability of distinguishablity. 99 % fidelity sounds better than ~ 5 % probability of distinguishablity. Useful connection to the Uhlmann-fidelity:

11 Performance measures for processes Mean state fidelity: Worst-case fidelity: Nice, intuitive measures but they are no metric, not stable,. Phys. Rev. A 71, (2005)

12 How-To State-Tomography

13 How to state tomography 3N measurement settings for an N-qubit system 8-qubit W state 6561 settings H. Häffner et at., Nature 438, (2005)

14 Characterize larger states? State Tomography Qubits Time for Data Time for Eval 1 10s 1s 4 4 min 5s h 1d h?? Use methods that require less data. Employ models and tools that fit into your memory. Run algorithms that generate a result in a reasonable amount of time. arxiv:

15 Process-Tomography Investigate how the complete Hilbert space evolves sample the complete Hilbert space as an input, investigate the output state for each input. Sampling the input Hilbert space: 0>, 1>, +x>, +y> Total number of measurements: 4N. 3N = 12N settings Largest system that has been investigated via full process tomography: 3-qubit systems (Toffoli, QFT,...)

16 Examples for processes Toffoli CCPhase T. Monz et al., PRL 102, (2009) M.Reed et al., Nature 482, (2012)

17 Compressed Sensing The purest state compatible with my (full-rank incomplete) data Parameters: r 2N (compressed sensing) vs 4N No need to assume rank done automatically Convex optimisation tons of (fast) routines Tight error bounds Directly applicable to processes Phys. Rev. Lett. 105, (2010) arxiv:

18 Permutationally invariant & Matrix product states Only ½ (N2 + 3N + 2) settings!!! Scalable ideas, but not (yet) extended to processes Phys. Rev. Lett. 105, (2010) Nat. Commun., 1 (9) 149

19 Permutationally invariant & Matrix product states Only ½ (N2 + 3N + 2) settings!!! Scalable ideas, but not (yet) extended to processes Phys. Rev. Lett. 105, (2010) Nat. Commun., 1 (9) 149

20 Why reconstruct processes? Fidelity Locality Markovianity Mean performance Worst-case performance... Cross talk Phase shifts Basis errors Possible correction pulses...

21 Why not to do process tomography? Markovianity & Concatenation of processes Numerical concatenation can be misleading in the non-markovian regime.

22 Why not to do process tomography? Spin echo cannot be described by concatenated processes Numerical concatenation can be misleading in the non-markovian regime.

23 Why not to do process tomography? In the lab it could be even worse

24 Why not to do process tomography? Noise spectrum does matter

25 Why not to do process tomography? Cross talk & embedding in larger registers A single-qubit operation looks good, but might show extensive cross-talk when embedded in a larger system.

26 Why not to do process tomography? Entire process = Initialisation Process Readout High-fidelity gates via randomized benchmarking Only derive the fidelity? Is that enough? New J. Phys (2009)

27 Direct Fidelity Estimation Fidelity = probability to detect the desired state So why not measure it directly? (1) Write down density matrices in terms of Pauli expectation values ( like a vector) (2) The fidelity is simply the scalar product of these two vectors. (3) Whatever Pauli expectation value is zero, you don't care about as it doesn't contribute less measurements Ideally measure the state directly, but DFE allows for an efficient decomposition into Pauli measurements. Phys. Rev. Lett. 106, (2011)

28 Key points of tomograhpies (1) Full tomography is hard and does not scale up. (2) There are efficient alternatives to full tomography. (3) If you are only interested in one parameter, measure it directly.

29 Experimental setup ~70 µm

30 Measuring the qubit Repeat times

31 How it really looks like

32 Qubits coupling to an oscillator 1> 1> > 0> n= 1> 0> n=1 n=0 n=2

33 Common motion of the ions in the trap Use common motion of the ions as a databus

34 Local operations Arbitrary local operations Feature: State and process tomography fully automated

35 Local operations

36 Local operations

37 Entangling operations

38 Entangling operations

39 An off-resonantly driven harmonic oscillator 11> n+1> n> n-1> n+1> 01> n> n-1> 00> n+1> n> 01> n-1> n+1> n> n-1> Multi-qubit interaction Combined with arbitrary local operations universal set of gates K. Mølmer, A. Sørensen, Phys. Rev. Lett. 82, 1971 (1999)

40 Generating large entangled states Ions 2 3 Pop., % 99.5 Coh., % Fid., % Single-shot 14-atom entanglement With a confidence of 76%!!

41 What can you do with this? Teleportation of a single qubit (2004) 8-qubit W state (2005) Toffoli gate (2009) Realize a bound entangled state (2010) 14-qubit GHZ state (2011) Repetitive quantum error correction (2011) Digital quantum simulation of open and closed systems (2011) Can we use our toolbox to enhance tomographies?

42 Direct characterization of quantum dynamics process on collective measurement qubits S system A ancilla input state prep. system qubit S S A A in the Bell basis ϵ S A entangle 4 pi, j=tr (Pi ϵ(ρ j ))= m, n=1 χm, n Λ im,, jn Choose input and measurement that inversion is possible i, j T Λ m,n =Tr (Pi (σ m I )ρ j (σ n I) ) 1 i, j m,n χi, j =(Λ ) pm,n M. Mohseni and D. A. Lidar, Phys. Rev. Lett. 105, (2006).

43 Direct characterization of quantum dynamics process on collective measurement qubits S system A ancilla input state prep. system qubit S S A A entangle ϵ in the Bell basis S A Pi={ ϕ ϕ, ψ ψ, ψ ψ, ϕ ϕ } ρ1 =( )/ 2 ρ2 =(α 11 +β 00 )/ 2 ρ3 =(α ++ x +β -- x )/ 2 Non maximally entangled states. ρ4 =(α ++ y +β -- y )/ 2 M. Mohseni and D. A. Lidar, Phys. Rev. Lett. 105, (2006).

44 Example of DCQD Let's make an example: Consider the first input state ρ1 =( )/ 2 A.) No process affecting S Identity: Tr(P i ϵ(ρ1)) ε= I ρ1 Input state ϵ( ρ1 )= ϕ + Output state I σx σy σz ( BSM I σx σy σz χ 11 χ 21 χ 31 χ 41 χ12 χ 22 χ 32 χ 42 χ13 χ23 χ33 χ 43 χ 14 χ 24 χ 34 χ 44 ) Tr( ϕ+ ϕ+ ρ1)=1 Tr( ψ+ ψ+ ρ1)=0 Tr( ψ- ψ- ρ1)=0 Tr ( ϕ- ϕ- ρ1 )=0

45 Example of DCQD B.) A phase-flip on the system qubit ( )/ 2 ρ1 =( )/ 2 ρ1 Input state ε= I Tr(P i ϵ(ρ1)) ϵ( ρ1 )= ϕ - Output state I σx σy σz ( BSM I σx σy σz χ 11 χ 21 χ 31 χ 41 χ12 χ 22 χ 32 χ 42 χ13 χ23 χ33 χ 43 χ 14 χ 24 χ 34 χ 44 ) Tr ( ϕ+ ϕ+ ρ1)=0 Tr ( ψ+ ψ+ ρ1)=0 Tr ( ψ- ψ- ρ1)=0 Tr ( ϕ- ϕ- ρ1 )=1

46 Experimental results ϵ Process DCQD, F(%) SQPT, F(%) Identity 97.5 ± ± 1.3 σx 96.5 ± ± 1.3 σy 96.6 ± ± ± ± 2.9 Phase damping 97.4 ± ± 0.8 Amplitude damping σ x rotation σ y rotation

47 Measuring non-unitary processes PROCESS AMPLITUDE DAMPING PHASE DAMPING EXPERIMENT IDEAL

48 Measuring a process with a single setting ϵ Expanding the ancilla Hilbert space Information on process encoded in ancillas by many-body interactions 16 Bell state operators and 1 input state

49 Measuring a process with a single setting ϵ Expanding the ancilla Hilbert space Information on process encoded in ancillas by many-body interactions 16 Bell state operators and 1 input state 4 pi=tr (Pi ϵ(ρ))= m, n=1 χm,n Λ im, n Choose input and measurement that inversion is possible

50 Measuring a process with a single setting ϵ Process DCQD, F(%) SQPT, F(%) Identity 99.7 ± ± 1.3 σx 97.3 ± ± 1.3 σy 99.8 ± ± 1.4 Measure input state and adapt theory

51 Characterization of real noise So far we discussed the full characterization of discrete processes acting on one qubit. Had to assume that process acting only on system qubit REAL: whole system is affected by noise (decoherence) Process of decoherence is dynamical Description of a single qubit interacting with the environment Dissipation and D Dissipation and Dephasing described by the dynamical parameters T 1 and T2 ( a b ρ= * b 1 a ) χ (t, T 1, T 2 ) ( t /T 1 (a 1/2)e ϵ(ρ)= * t / T b e 2 t /T 2 be t / T 1 (a 1/2)e 1/2 1 )

52 Preparation in the state: ρ1 =( )/ 2 Process applied on both qubits: χ 1,1 χ 4,4 =e 2t / T 2 1 2(χ 2,2 +χ 3,3 )=e Relaxation times depend only on populations One measurement setting required No full QPT partial information 2t/ T 1

53 Characterize relaxation times Characterize Dephasing and spontaneous decay on both qubits Input state: ρ1 =( )/ 2 Phase flip transfers into Bit flip transfers into ( )/ 2 ( )/ 2 ( )/ 2

54 Characterize quantum dynamics 1 S S S S 1 A A A A ρ1 =( )/ 2 Waiting time t Populations are measured for each time step t Each experiment was repeated 150 times BSM

55 Characterize quantum dynamics T 2=19.4 (8)ms Ramsey contrast experiment T 1=1170( 26)ms Spontaneous decay measurement T DCRT =18.8 (5) ms 2 T DCRT 1 =1160 (17) ms Input state ρ1 =( )/ 2

56 Conclusion and Outlook QPT scheme using entangled input states + measurement in entangled basis 4 measurement settings Coherent, amplitude- and Phase-damping processes with high fidelities Direct characterization of dynamical processes: Only partial information necessary Efficient noise characterization One measurement setting Powerful tool to efficiently quantify Hamiltonian parameters 3 assume that Hamiltonian is known Demonstration of QPT with one measurement setting M. Mohseni et al., Phys. Rev. Lett. 77, (2008). 3

57 Summary Direct characterization of quantum dynamics + Reduces number of measurement settings + Allows the characterization of a process with a single shot + Allows to determine the parameters of a quantum process - Is not scalable for arbitrary processes D. Nigg et al., arxiv: See also: T. Graham et al., arxiv:

58 Most metrics require ρ or χ 3 of the DiVincenzo Criteria: Initialize your qubits (Universal) set of quantum gates State readout When you can perform all of these three points, you ought to be able to do state and process tomography.

59 Model assumptions for scalability (state tomo methods) Compressed sensing Permutationally invariant (PI): GHZ, W, Dicke, Matrix-product states (MPS): generated/modeled by local interaction Multi-scale entanglement renorm. ansatz (MERA) more than close neighbour interaction

60 Hofmann bounds MS (3 qubits): 91% < Fproc < 94% (1) Look at the mean fidelity of two complementary sets of quantum states (2) Measure the output state fidelities, calculate the mean for each set Bound pretty much useless for bad gates Phys. Rev. Lett. 94, (2005)

61 Validity Positivity I. Take first half of the data II. Make a linear reconstruction III. Find most-negative eigenvalue and corresponding state. IV. Do MLE on the second half of the data and calculate the expectation value. V. Employ Hoeffding's tail inequality to check trust Linear Dependencies I. Take first half of the data II. Make a linear reconstruction III. Define witness IV. Do MLE on the second half of the data and calculate the expectation value. V. Employ Hoeffding's tail inequality to check trust arxiv:

62 The international Team 2012 FWF SFB Industrie Tirol AQUTE $ IQI GmbH