Power-efficient production of photon pairs in a tapered chalcogenide microwire

0.23 MB·English

by  Evan Meyer-Scott

Power-efficient production of photon pairs in a tapered chalcogenide microwire
Evan Meyer-Scott,1,∗ Audrey Dot,1 Raja Ahmad,2 Lizhu Li,2 Martin Rochette,2 and Thomas Jennewein1,3
1Institute for Quantum Computing and Department of Physics and Astronomy,
University of Waterloo, 200 University Ave W, Waterloo, Ontario, Canada N2L 3G1
2Department of Electrical and Computer Engineering, McGill University,
3480 University Street, Montr´eal, Qu´ebec, Canada H3A 2A7
3Quantum Information Science Program, Canadian Institute for Advanced Research, Toronto, ON, Canada
Using tapered fibers of As2Se3 chalcogenide glass, we produce photon pairs at telecommunica-
tion wavelengths with low pump powers. We found maximum coincidences-to-accidentals ratios of
2.13±0.07 for degeneratepumpingwith 3.2µWaverage power, and1.33±0.03 for non-degenerate
5 pumping with 1.0µW and 1.5µW average power of the two pumps. Our results show that the
1 ultrahigh nonlinearity in these microwires could allow single-photon pumping to produce photon
0 pairs, enabling the production of large entangled states, heralding of single photons after lossy
2 transmission, and photonic quantuminformation processing with nonlinear optics.

b
e
F Photon pair sources are the simplest and most abun- from low-power non-degenerate pumps, as would be re-
dantdevices usedto produce quantum entanglement[1], quired for a single-photon-pumping experiment.
5
and are normally based on nonlinear optical effects that Our microwires combine a high nonlinear coeffi-
2
requirelargeinputlightintensities. Withhighlyefficient cient [18] (n = 1.1×10−17m2/W) with low cross-
2
] photon sources, it becomes possible to use spontaneous sectional area (0.24µm2) to produce a large waveguide
h parametricdown-conversion(SPDC)orfour-wavemixing nonlinearparameter(γ =188W−1m−1),∼100000times
p
(FWM)actingonsinglephotonsaspumpsinlaterstages larger than standard silica fibers, and ten times larger
-
t of quantum information experiments [2, 3], enabling ad- than As S chalcogenide waveguides [10]. Furthermore,
n 2 3
vanced protocols like loophole-free Bell tests via photon in contrast to on-chip waveguides, microwires are drop-
a
u heralding [4] and photonic quantum computing [5]. in compatible with existing single-mode silica fiber, are
q However,theconversionefficiencyofonepumpphoton madewithlengthsuptotensofcentimeters,anddemon-
[ intoapairthroughSPDCinχ(2) mediahasnotsubstan- strate low loss. Coupling to the microwire is accom-
tially increased [6, 7] beyond 10−6. Focus has shifted plished by gluing standard single-mode fiber (SMF) to
2
to developing efficient FWM in χ(3) fibers and waveg- the chalcogenide step-index fibers on either side of the
v
2 uides, using materials with large nonlinearities such as tapered region. A coating of poly(methyl methacrylate)
3 silicon[8]andchalcogenideglasses[9,10],andalsousing (PMMA) increases the mechanical robustness and influ-
5 resonator-enhancedprocesses[11,12],atomicvapors[13], ences the phasematching properties of the microwires.
3 and microstructured silica fibers [5]. These would allow Broadband phasematching of up to 190nm bandwidth
0
non-degeneratepumpingwithasinglephotonandstrong attelecommunicationwavelengthshasbeen shownusing
.
1 pump to produce pairs. classical four-wave mixing [19].
0 Unfortunately, noneofthese sourcesis totally suitable
v:15 fcfiohcraielcnleocnnygveewrtittoihndgceossniingvgnelnetihepenhtdooetovpniecsreatttoihoapntaaiornffsd,eralsonwtdhenitohirsiegemh.eaSsintilsiecfaa- er out ofple (nW)0.51 (a) P1 CMaelacsuularteiomnent
Xi devices cannot reach sufficiently high efficiency due to Powsam 0
low nonlinearity [14], and though silicon exhibits much 1535 1540 1545 1550 1555 1560 1565
ar lower noise [15], it suffers from two-photon and free- Wavelength (nm)
carrier absorption [16], limiting the maximum useable ofW)4 (b)
ponuamtporspoowruers.ingEanthoamncicinvgaptohresnreoqnuliinreesartihtyeitnhpruotugphhorteosn- er out ple (n2
wm P1 P2
tobenarrowband,limitingthetypesofinitialsourcesfor Posa0
1540 1545 1550 1555 1560 1565 1570
this photon.
Wavelength (nm)
Here we demonstrate production of photon pairs in
tapered As Se microwires, capable in principle of con-
2 3 FIG.1. (Coloronline)Phasematchingobtainedinthechalco-
verting a broadband single photon into a pair with 10−3 genidemicrowire(redmarkers)for(a)degeneratepumpingat
probability[17]. Firstwecharacterizethefour-wavemix- 1548.5nmand(b)non-degeneratepumpingat1549.0nmand
ingpropertiesofthemicrowires,thenweproducephoton 1562.1nm. ThepumpsP1andP2aremarkedwiththickblue
pairsfromdegeneratepumping,andfinallyproducepairs bars,and thephasematchingcalculated directly from sample
parameters is theblack lines.
The tapered fiber used in this work has a 12cm long
∗ [email protected] microwire with a diameter of 550nm, with total inser-
2
tion loss including pigtails of 10dB. We estimate the PC
P1 PBS PC
breakdownof the losses as follows: inside the microwire, OPO P2 DWDM
5dBduetosidewallroughness,absorptionoftheevanes- As2Se3 fiber
cent field by the polymer coating, and mode conversion PMMA
12 cm
in the tapering sections; at the SMF-chalcogenide inter-
faces, 0.5dB per interface due to Fresnel reflection, and NFAD AWG Microwire
2dB per interface due to mode mismatch because the detector DWDM DWDM
Trigger AWG
glued pigtails have been disrupted by transport. Losses
DWDM
pbcreeoFrewniSgirrMueerpFeao-t1cr∼thesa1dhl5co[5ow20g0sen].ntmihd,eefapirnhtafersroefmmacaeatncayhsiznleogrwoprdaosisfip0lee.6ro6sifodonBuwrhamavveie-- Timetag deItDeQct or Power out ofsample (dBm)----876510000450 1W50a0vel1e5n5g0th (1n6m0)0 1650
length, for a degenerate pump (a) and for two non-
degenerate pumps 13nm apart (b). These data were FIG.2. (Coloronline) Experimentalsetupforthegeneration
obtained by pumping and seeding with continuous-wave ofphotonpairswithachalcogenidemicrowirepumpedbyone
lasers with 190µW coupled power (inferred power inside (P1, black) or two (P1 and P2, grey) beams from an optical
the microwire after accounting for coupling losses) each. parametric oscillator (OPO). For non-degenerate pumping,
the two pumps were made to be copolarized with polariza-
For the degenerate case, the seed laser was scanned to
tion controllers (PC) and a fiber polarization beamsplitter
the shorter wavelength side of the pump, and for the
(PBS). In both degenerate and non-degenerate cases a final
non-degenerate case, the seed was scanned between the
PC sets the polarization entering the microwire. The pump
twopumpwavelengths,leavingagapinthemiddlewhere
and signal/idler beams were filtered by dense wave-division
the seed and signal cross over. In both cases the output
multiplexers (DWDM) and an arrayed waveguide grating
signal was filtered through a dense-wave division multi- (AWG). The signal photon was detected by a free-running
plexer (DWDM) and measured on a power meter. The negative-feedbackavalanchephotodiode(NFAD)whichgated
valuesreportedherehavetheDWDMlossesfactoredout. the idler’s single-photon detector (IDQ). Both signal and
We calculated the expected phasematching and power idlerdetectionsignalswererecordedandtimestampedbythe
timetag unit from Universal Quantum Devices, Inc. Inset:
outputsfromastandardfour-wavemixingtreatment[21],
Ramanscatteringnoisespectrumofthemicrowirepumpedat
with the length, diameter, insertion loss, and nonlinear
awavelengthof1550nm. Twominimaoccurat±40nmfrom
parameter as mentioned above, and calculated loss in
the pump [18]. This spectrum includes a 10nm-wide notch
the microwire [22] of 5.1dB/m. In order to find the
filterwith 30dBblockingcentered at 1550nm. Thepumpat
propagation constant and effective refractive index in- 1550nm extends off the top of the graph to −34 dBm, and
side the microwire, we solved the characteristic equa- ourspectrometer’s sensitivity is limited to −80 dBm.
tion for a step-index fiber with As Se core [18] and
2 3
PMMA cladding [23]. Even without fitting parameters,
the phasematching calculations in Fig. 1 agree with the delays. Atthelowestpower,detectordarkcountsatran-
measured data. domtimesbecomeprominent,whileatthehighestpower,
Next we produced photon pairs using a degenerate accidentalcoincidences frommultiple photonpairsmake
pump, with the setup shownin Fig. 2. The output of an a large contribution to the noise. When the idler chan-
optical parametric oscillator at 1553.33nm with ∼4ps nel is moved to 1558.17nm (black lines in Fig. 3) such
pulse length and 76MHz repetition rate was filtered that photon pairs detected would not conserve energy
through a DWDM and sent through the tapered chalco- and could not be from four-wave mixing, the peak at
genide microwire. The FWM output signal and idler 22ns falls to the same height as the others.
photons were split into the 1550.12nm and 1556.56nm As seen in Fig. 4(a), the number of pairs per pulse in-
channels of a DWDM, and subsequently filtered in an creasesquadraticallywithpumppower,withdropoffseen
AWG and DWDM respectively, giving total pump iso- at high count rates due to the dead time of the NFAD
lation of 118dB and 122dB. The signal photon was detector. Tocomparewithpreviousphotonpairsources,
detected with a free-running InGaAs negative-feedback we calculate the number of pairs produced per second,
avalanche photodiode (NFAD) with 10% detection effi- per nanometer of signal bandwidth, per milliwatt of av-
ciency and 100 dark counts per second [24], which gated erage pump power as 2.5×108pairs/s/nm/mW for the
anid201InGaAssinglephotondetectorfromIDQforthe datapointwith30µWpumppower. Thisissignificantly
idler photon, with gate width 50ns and 20% efficiency. abovewhatispossibleinχ(2) crystals[25],andrivalsthe
The gate out and detector channels from the id201 were highest values reported in silicon [26]. Note that our fil-
recorded in a time-tagger to either produce timing his- ters had 0.5nm bandwidth, and milliwatt pump powers
tograms between signal and idler, or to filter with a 2ns are too large to avoid damage to the microwire.
timing window to record coincident counts. In Fig. 4(b), the coincidences-to-accidentals ratio
The timing histograms in Fig. 3 show the presence of reaches its maximum value of 2.13 ± 0.07 at a cou-
photon pairs, as evidenced by a higher peak at the time pled pump power of 3.2µW. This ratio is defined as
delay of 22ns than the background peaks at other time CAR=C/A, where C is the total number of coincident
3
Coupled peak pump power (mW)
FWM, 49 µW 0 5 10 15
FWM, 20 µW se0.06
nt counts01 FFNWWoiMsMe,, o22n.55l0y µ nWW b. per pul0.04 (aSM)iemausluartieomnent
oincide01 Pair pro0.020
c
d 2
e1
aliz AR1.5
m0 C 1 (b)
Nor 0.5
1
0
0 5 10 15 20 25 30
0 Coupled average pump power (µW)
0 10 20 30 40
Offset time (ns)
FIG.4. (Coloronline)Pairprobabilityperlaserpulsewithac-
cidentalssubtracted(a) andcoincidences-to-accidentals ratio
FIG.3. (Coloronline)Timinghistogramsforphotonpairpro-
ductionforcoupledaveragepumppowersof250nW,2.5µW, (b)asmeasuredfordegenerate-pumpingFWMinourchalco-
20µW, and 49µW. The photon pairs appear in the peak genide microwire. The curves are from a FWM simulation
at thetime offset between signal and idlerdetection of22ns, includingmeasured backgroundcounts,with thepumppulse
whichishigherthantheaccidentalcoincidencepeaksatother length inside themicrowire as a fitting parameter. CAR>1
indicates photon pairs are detected above the noise. Error
delays. In all cases theaccidentals peak height is normalized
bars are based on poissonian uncertainty in photon counts,
to1. Forthe“noiseonly”measurement(blacklines),thefilter
and are smaller than symbol size for (a). The x-axes on fig-
channels connected to the detectors did not conserve energy
ures (a) and (b) coincide and show both peak and average
withthepump,leadingtothedisappearanceofthelegitimate
photon pairs. The double peak at 35ns could be due to the power.
timing electronics in the IDQ detector or timetagger, but no
eventsare lost.
CAR = 1.33±0.03
s Copolarized pumps
nt
u
counts in the main coincidence peak and A is the num- nt coCrosspolarized pumps CAR = 1.101±0.009
ber of accidental coincidences, and has a lower bound of e
d
CAR = 1 for no timing correlation. At a coupled pump nci CAR = 0.97±0.03
power of 490nW, the CAR was 1.5±0.2, and statistical coi Bad pump delay
d
sthigenicfiocinancicdeeinnccerseaasreedcwoliltehctiendcraetastihneg2p2unmspopffosewtert.imHeeroef malize Only P1 CAR = 0.97±0.05
Fig. 3, while the accidentals, which are due to detector or
N CAR = 1.00±0.05
dark counts, double-pair emissions from FWM, and Ra-
Only P2
man and other optical noise, are collected at 9ns, which
25 30 35 40 45 50 55
allows an estimation of the contribution of accidentals Offset time (ns)
to the main coincidence peak. At low power, the CAR
decreases due to the small number of real photon pairs FIG. 5. (Color online) Timing histograms for photon pair
comparedto noisephotons,andathighpower,the CAR production for non-degenerate pumping. The photon pairs
appearin thepeakat timingdelay43nsonlyforthetoptwo
slowly decreases due to double-pair emissions from four-
cases,withcopolarizedandcrosspolarizedpumpsrespectively,
wavemixing. ThefitstopairprobabilityandCARcome
giving CAR > 1. In the other cases, where FWM is not
from the FWM calculation described above, which feeds expected, the peak at 43ns vanishes and CAR = 1 within
into a quantum-optical simulation including FWM, loss,
error. Hereeachdatasethashadthebackground“noiseonly”
detector models, and background counts measured with countssubtracted.
a continuous-wave pump. Here the pump pulse length
inside the microwire is used as a fitting parameter and
found to be 25ps. in the microwire as in the grey lines and text in Fig. 2.
In order to approach the regime of converting a sin- To find evidence of photon pairs with non-degenerate
gle photon into a pair, the two pump photons absorbed pumping, we took timing histograms as shown in Fig. 5.
in the FWM process must be non-degenerate in wave- Wepresentanumberofdifferentscenarios: co-andcross-
length. Since FWM probability goes as the product of polarized pumps, changing the delay between the two
the pump peak powers, it is important that both pumps pumps so they do not overlap in the microwire, and re-
be synchronously pulsed, rather than continuous-wave. moving either of the pumps. Only the first two cases,
To demonstrate non-degenerate pumping, we took two withthecrosspolarizedpumpsproducing4/9thenumber
outputsfromanopticalparametricoscillator,passingone of pairs of the copolarizedpumps [21], provide a peak at
through a trombone delay line to synchronize the pulses the proper delay after background subtraction, indicat-
4
5000 into two will require a single-photon pump of 10pW av-
Coinc. measurement
c. andcounts44050000 ACAccoccin.. cms.i mefitauslautrieomnent einraFgieg.p1o,wwere.aDreuceutrorenthtleynfoarrcroedwtpohcaoslelmecattpchhiontgonshpoawirns
Coinacc. 3500 itnamainwaatveedlewnigtthhnroeigseio(nincsleotseintoFitgh.e2)p.uTmhpeslothwaetstispucomnp-
3000
50 60 70 80 90 100 110 120 130 power for non-degenerate pumping with which we found
Relative pump delay (ps) CAR>1 was 480nW. Additionally, reaching 10−3 con-
versionefficiency[17]wouldrequireanaveragepowerfor
FIG. 6. (Color online) Coincident and accidental counts
the other pump of 130µW, whereas we found CAR > 1
versus time delay between pump pulses for non-degenerate
for non-degenerate pump power only up to 3µW. This
pumping. Photon pairs are only produced when the pumps
does not mean that the high conversion efficiency can-
overlap in the microwire; accidentals measured 13ns (one
not be achieved, but it requires moving the signal and
pump period) later also increase as the pumps overlap due
to multi-pairemissions. Here theaccidentals are higher than idler much farther in wavelength from the strong pump
thecoincidencesoutsidethepeakbecausetheaccidentalswere toavoidnoiseandbroadeningtheirphasematchingband-
collectedfromamoreefficientpartoftheIDQdetector’sgate. widthstoincreaseefficiency. Bycarefullycontrollingcore
Thelineonthecoincidencedataisagaussian fitwhichisfed sizeandcladdingmaterial,itwillbepossibleinfutureto
into a FWM simulation to generate theaccidentals curve. fabricatemicrowireswithengineereddispersiontorepro-
ducebroadbandphasematchingpreviouslyobtained[19].
Ifthecurrent0.5nmphotonbandwidthcanbeexpanded
to 40nm, centered 100nm from the pump, 80 times
ingthesephotonsareduetonon-degenerateFWMasde-
moreefficientconversionisexpected. Thesephotonpairs
sired. The maximum CAR for non-degenerate pumping
was 1.33±0.03 with 1.0µW and 1.5µW coupled powers would be produced in the Raman gain dips or even be-
yond the gain peaks, which would improve CAR drasti-
of the two pumps, while the CAR for the lowest asym-
metric pump power was 1.17±0.06 with coupled pump cally as shown in chalcogenide waveguides [27, 28].
We have presented evidence of photon pairs produced
powers of 480nW and 1.5µW.
in an As Se chalcogenide microwire for both degener-
2 3
Unfortunately due to bad phasematching (see Fig. 1)
ate and non-degenerate pumping. Because this device
and the extra noise broughtby having two pumps, these
is a tapered fibre, the coupling of optical signals in and
data are not as clear as the degenerate case. A more
outisstraightforwardandstablewith roomforimprove-
convincing measurement is one of coincidences and acci-
ment in efficiency, making the system very interesting
dentalsversustime delaybetweenthe twopumps, where
for future applications. Through timing analysis of pho-
photon pairs are produced only when the two pumps
ton pairs, we found that the coincidences-to-accidentals
overlap, on top a constant background caused by spon-
ratio maintains a value of CAR > 1 over a wide range
taneousRamanscattering. InFig.6,weshowthesedata
of pump powers, and inferred a maximum pair produc-
for P1 = 1551.72nm, P2 = 1561.42nm, and the signal
tion rate inside the microwire (limited by detector dead
and idler wavelengths 1554.13nm and 1558.98nm. The time) of 2.9×106pairs/s. We look forward to reducing
coincidence curve is a gaussian fit to the data, which
the background noise by engineering the phasematching
is fed into a quantum-optical FWM simulation that in- conditions and pushing closer to 10−3 efficiency for con-
cludes higher-order emission but no other nonlinearities
verting a single photon into a pair.
to find the expected number of accidentals. The simu-
We acknowledge support from NSERC, Ontario Min-
latedaccidentalsagreewiththedata,indicatingthatthe
istry of Research and Innovation, CIFAR, FedDev On-
increased accidentals when the pumps overlap are due
tario, Industry Canada, and CFI. We are grateful to
only to photon emission statistics of the FWM process.
CoractiveHigh-Techforprovidingthechalcogenideglass
To achieve in future the splitting of a single photon used in the experiments.
[1] R. Horodecki, P. Horodecki, M. Horodecki, and [5] N.K.Langford, S.Ramelow, R.Prevedel,W.J. Munro,
K.Horodecki, Rev. Mod. Phys. 81, 865 (2009). G.J.Milburn, andA.Zeilinger,Nature478, 360 (2011).
[2] T. Guerreiro, A. Martin, B. Sanguinetti, J. S. [6] S. Tanzilli, H. de Riedmatten, H. Tittel, H. Zbinden,
Pelc, C. Langrock, M. M. Fejer, N. Gisin, P.Baldi,M.DeMicheli,D.B.Ostrowsky, andN.Gisin,
H. Zbinden, N. Sangouard, and R. T. Thew, Electron. Lett. 37, 26 (2001).
Phys.Rev.Lett. 113, 173601 (2014). [7] Q. Zhang, X. Xie, H. Takesue, S. W. Nam,
[3] D.R.Hamel,L.K.Shalm,H.Hu¨bel,A.J.Miller,F.Mar- C. Langrock, M. M. Fejer, and Y. Yamamoto,
sili, V. B. Verma, R. P. Mirin, S. W. Nam, K. J. Resch, Opt. Express15, 10288 (2007).
and T. Jennewein, NaturePhoton. 8, 801 (2014). [8] H. Takesue, IEEE J. Sel. Top. Quantum Electron. 18,
[4] A. Cabello and F. Sciarrino, Phys. Rev. X 2, 021010 1722 (2012).
(2012). [9] C. Xiong, L. G. Helt, A. C. Judge, G. D. Mar-
5
shall, M. J. Steel, J. E. Sipe, and B. J. Eggleton, [19] R. Ahmad and M. Rochette, Opt. Express 20, 9572
Opt.Express 18, 16206 (2010). (2012).
[10] C. Xiong, G. D. Marshall, A. Peruzzo, M. Lobino, A. S. [20] C. Baker and M. Rochette, Photonics Journal, IEEE,
Clark,D.-Y.Choi,S.J.Madden,C.M.Natarajan,M.G. Photonics Journal, IEEE 4, 960 (2012).
Tanner, R. H. Hadfield, S. N. Dorenbos, T. Zijlstra, [21] G. P. Agrawal, Nonlinear Fiber Optics, 5th ed. (Aca-
V. Zwiller, M. G. Thompson, J. G. Rarity, M. J. Steel, demic Press, Boston, 2013).
B. Luther-Davies, B. J. Eggleton, and J. L. O’Brien, [22] C. Baker and M. Rochette, Opt. Express 18, 12391
Appl.Phys.Lett. 98, 051101 (2011). (2010).
[11] J. Chen, Z. H. Levine, J. Fan, and A. L. Migdall, [23] S.N.Kasarova,N.G.Sultanova,C.D.Ivanov, andI.D.
Opt.Express 19, 1470 (2011). Nikolov, Opt.Mater. 29, 1481 (2007).
[12] C. Reimer,L. Caspani, M. Clerici, M. Ferrera, M. Kues, [24] Z. Yan, D. R. Hamel, A. K. Heinrichs, X. Jiang, M. A.
M. Peccianti, A. Pasquazi, L. Razzari, B. E. Lit- Itzler, andT. Jennewein, Rev.Sci. Instrum.83, 073105
tle, S. T. Chu, D. J. Moss, and R. Morandotti, (2012).
Opt.Express 22, 6535 (2014). [25] F. Steinlechner, S. Ramelow, M. Jofre, M. Gilaberte,
[13] C.-K. Chiu, Y.-H. Chen, Y.-C. Chen, I. A. Yu, Y.-C. T. Jennewein, J. P. Torres, M. W. Mitchell, and
Chen, andY.-F.Chen,Phys. Rev.A 89, 023839 (2014). V. Pruneri, Opt.Express 21, 11943 (2013).
[14] M.A.Foster,A.C.Turner,M.Lipson, andA.L.Gaeta, [26] E. Engin, D. Bonneau, C. M. Natarajan, A. S.
Opt.Express 16, 1300 (2008). Clark, M. G. Tanner, R. H. Hadfield, S. N. Doren-
[15] K. Harada, H. Takesue, H. Fukuda, T. Tsuchizawa, bos, V. Zwiller, K. Ohira, N. Suzuki, H. Yoshida,
T. Watanabe, K. Yamada, Y. Tokura, and S. Itabashi, N.Iizuka,M.Ezaki,J.L.O’Brien, andM.G.Thompson,
IEEE J. Sel. Top. Quantum Electron. 16, 325 (2010). Opt. Express21, 27826 (2013).
[16] B. J. Eggleton, B. Luther-Davies, and K. Richardson, [27] J. He, C. Xiong, A. S. Clark, M. J. Collins, X. Gai, D.-
NaturePhoton. 5, 141 (2011). Y. Choi, S. J. Madden, B. Luther-Davies, and B. J.
[17] A. Dot, E. Meyer-Scott, R. Ahmad, M. Rochette, and Eggleton, J. Appl.Phys. 112, 123101 (2012).
T. Jennewein, Phys.Rev.A 90, 043808 (2014). [28] M. J. Collins, A. S. Clark, J. He, D.-Y. Choi, R. J.
[18] R. E. Slusher, G. Lenz, J. Hodelin, J. Sanghera, L. B. Williams, A. C. Judge, S. J. Madden, M. J. Withford,
Shaw, andI.D.Aggarwal, J.Opt.Soc.Am.B21,1146 M.J.Steel,B.Luther-Davies,C.Xiong, andB.J.Eggle-
(2004). ton, Opt.Lett. 37, 3393 (2012).