Intermittent photocatalytic activity of singleCdS nanoparticlesYimin Fanga, Zhimin Lia, Yingyan Jianga, Xian Wanga, Hong-Yuan Chena, Nongjian Taoa,b, and Wei Wanga,1

aState Key Laboratory of Analytical Chemistry for Life Science, School of Chemistry and Chemical Engineering, Nanjing University, Nanjing 210093, China;and bCenter for Bioelectronics and Biosensors, Biodesign Institute, Arizona State University, Tempe, AZ 85287

Edited by Thomas E. Mallouk, The Pennsylvania State University, University Park, PA, and approved August 18, 2017 (received for review May 24, 2017)

Semiconductor photocatalysis holds promising keys to addressvarious energy and environmental challenges. Most studies todate are based on ensemble analysis, which may mask criticalphotocatalytic kinetics in single nanocatalysts. Here we report astudy of imaging photocatalytic hydrogen production of singleCdS nanoparticles with a plasmonic microscopy in an in operandomanner. Surprisingly, we find that the photocatalytic reactionswitches on and off stochastically despite the fact that the illumi-nation is kept constant. The on and off states follow truncated andfull-scale power-law distributions in broad time scales spanning3–4 orders of magnitude, respectively, which can be describedwith a statistical model involving stochastic reactions rates at mul-tiple active sites. This phenomenon is analogous to fluorescencephotoblinking, but the underlying mechanism is different. As in-dividual nanocatalyst represents the elementary photocatalyticplatform, the discovery of the intermittent nature of the photo-catalysis provides insights into the fundamental photochemistryand photophysics of semiconductor nanomaterials, which is antic-ipated to substantially benefit broad application fields such asclean energy, pollution treatment, and chemical synthesis.

single-nanoparticle catalysis | intermittent activity | semiconductorphotophysics | semiconductor photochemistry | surface plasmonresonance microscopy

Photocatalytic H2 production based on semiconductor nano-catalysts is one of the most promising solutions for the energy

crisis via solar-to-chemical energy conversion (1–3). In recentyears, single-nanoparticle catalysis, which studies the catalytic ac-tivity at single-nanocatalyst level, has received great attention dueto its unique strengths to understand the microscopic catalytic ki-netics and mechanism (4–6). For semiconductor photocatalysts,single-nanoparticle photocatalysis is particularly interestingbecause of their comprehensive photophysical processes in-cluding photoinduced carrier generation, migration, and re-combination. For example, the fluorescence photoblinkingof individual semiconductor quantum dots has been a long-standing mystery since its discovery two decades ago (7), andits underlying mechanism remains an open question (8–10).When considering the fact that the photocatalytic activity ofsemiconductor nanomaterials is intrinsically associated withtheir photophysical processes, it naturally raises a question ofwhether the photocatalytic activity is also intermittent at single-nanoparticle level, or is it simply constant or monotonicallydecaying (due to photocorrosion and photopassivation), as onecould intuitively predict.This question has not been answered by the existing techniques for

studying single-nanoparticle catalysis such as single-molecule fluo-rescence (4–6) and dark-field scattering (11–13). Although it is ap-plicable to image individual dielectric nanoparticles with dark-fieldmicroscope (14), most studies so far have been focusing on plasmonicnanomaterials due to their relatively large scattering cross-section andthe sensitive spectral dependence on the surrounding environment.Single-molecule fluorescence microscopy is one of the most adoptedtechniques in this field by monitoring the counts and locations offluorescence bursts in a fluorogenic reaction, which indicate the

formation of fluorescent product molecules (15). Although powerful,fluorescence microscopy requires a fluorogenic model reaction andmight compromise the nature of many important reactions whereproducts are nonfluorescent (for example, photocatalytic H2 pro-duction reactions). Besides, because it often takes tens of minutes toaccumulate enough counts for statistical analysis, it has been mostlyused to study the spatial heterogeneity at a cost of temporal resolu-tion (4–6, 16). A previous study reported the inhibition and reap-pearance of photocatalytic activity of single Sb-doped TiO2 nanorodsas a result of the absorption and desorption of surface adsorbates(17). Analysis of the burst correlation times have also suggested theactivity fluctuations of individual gold (5) and platinum (18) nano-particles due to the dynamic surface restructuring of metal atoms.However, intermittent photochemical activity (i.e., stochastic blinkingbetween ON and OFF states) of single semiconductor nanoparticlesthat is regulated by its intrinsic photophysical processes has beenlargely unexplored.Here we use a surface plasmon resonance microscopy (SPRM) to

continuously monitor the H2 production rate of individual CdSnanoparticles in an in operandomanner, and report the discovery ofintermittent photocatalytic activity of single CdS nanocatalyst.SPRM is an optical microscopy that we recently developed to imagethe local refractive index (RI) distribution with a temporal resolu-tion up to microseconds and a spatial resolution around the dif-fraction limit (19–21). SPRM is capable of monitoring local H2concentration due to the large difference in the RI between H2Oand H2. Instead of being constant or monotonically changing, it wassurprisingly found that the photoinduced H2 production rate of

Significance

Semiconductor photocatalysis holds promising keys to addressvarious energy and environmental challenges. While conven-tional wisdom suggests a continuous photocatalytic reactionunder constant light illumination, in the present article we reportthe discovery of intermittent photocatalytic activity at single CdSnanoparticle level. The observed intermittent photocatalysis is aphotochemical consequence of its intrinsic photoexcitation pro-cesses. The latter is also responsible for the well-known fluores-cence photoblinking of single-semiconductor quantum dots, aphotophysical phenomenon that was discovered in the 1990s.The intermittent photocatalysis (a photochemical process) repor-ted here could be an exciting complement of the beautiful pictureof semiconductor photophysics and photochemistry, with signif-icant implications in many application fields from clean energy topollution treatment.

Author contributions: Y.F., H.-Y.C., and W.W. designed research; Y.F., Z.L., Y.J., and X.W.performed research; Y.F., Y.J., and W.W. analyzed data; Y.F., N.T., and W.W. wrote thepaper; and W.W. conceived and supervised the research.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1To whom correspondence should be addressed. Email: wei.wang@nju.edu.cn.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1708617114/-/DCSupplemental.

10566–10571 | PNAS | October 3, 2017 | vol. 114 | no. 40 www.pnas.org/cgi/doi/10.1073/pnas.1708617114

Dow

nloa

ded

by g

uest

on

July

25,

202

0

single CdS nanocatalyst switched between active (ON) and inactive(OFF) states frequently and alternatively. While OFF event followsa full-scale power-law distribution, ON event exhibits a truncatedpower-law distribution. The location of truncation depends on themaximum interfacial electron transfer rate (ket) occurring at thesingle nanoparticle. The remarkable intermittency of photocatalysisand its statistical characteristics are well modeled by hypothesizingmultiple reaction sites with stochastically distributed ket on each site.We first demonstrated the principle of SPRM for monitoring the

photocatalytic hydrogen production at individual CdS nanoparticles(Fig. 1A). These CdS nanoparticles had a diameter of ∼100 nm,which were synthesized and characterized according to the proce-dures described in SI Appendix, Figs. S1–S3. Subsequently, theywere deposited onto a gold-coated glass coverslip for SPRM im-aging (SI Appendix, Figs. S2 and S3). The gold film was passivatedby a self-assembled monolayer of HS–(CH2)17–CH3 molecules be-fore CdS deposition, to block the possible electron transfer betweenCdS nanoparticles and gold film (22, 23) (SI Appendix, Figs. S4 and

S5). The root-mean-square roughness of gold film was determinedto be ∼0.4 nm, which showed negligible interference to the surfaceplasmon polaritons (24). The coverage of the nanoparticles wastuned to be sufficiently low to minimize interference in the photo-catalytic reactions between adjacent nanoparticles. Light withwavelength of 680 nm (2.5 mW·mm−2) was used to excite planarsurface plasmons on the gold film with an objective-based totalinternal reflection configuration (19–21). Light reflected from thegold film was collected via the same objective and directed into acamera to form an SPRM image. A typical SPRM image of sevenCdS nanoparticles is shown in Fig. 1A, which reveals each nano-particle as a bright spot with a long parabolic tail. This parabolic tailis due to scattering of the surface plasmonic waves by the nano-particle that has a diameter smaller than the diffraction limit (21,25). The CdS nanoparticle-covered gold film was illuminated fromthe top with a blue light (λ < 500 nm, 14 mW·mm−2) to generatephotocatalytic reactions at the CdS nanoparticles. Hydrogen pro-duced by the reaction reduced the local RI surrounding each CdS

007300630053

photocatalysis ON

photocatalysis OFF

1 10 10010-4

10-3

10-2

10-1

100

101

1 10 100 100010-5

10-4

10-3

10-2

10-1

100

101Pr

obab

ility

den

sity

(1/s

)

ON duration time (s)Pr

obab

ility

den

sity

(1/s

)OFF duration time (s)

-10

0

10

20

0 2000 4000 6000

SPR

Inte

nsity

(mDe

g)

Time (s)

Time (s)

NP1 NP2 Background

Camera

Objective

Background

NP1NP2

Blue light(for photochemistry)

Red light(for SPR)

High-passfilter

H2H+

�me

SPR

Inte

nsity

Blue light ON

S2-

S22-h+

e-

Recomb. Ex.

Gold-coated coverslip with C18 SAM

A

B

C

Fig. 1. (A) Schematic illustration of SPRM setup for monitoring the H2 production rate on single CdS NPs. Upon the illumination of blue light, photoexcitedelectrons and holes are separated and move to different sites to reduce protons to generate H2 and to oxidize S2− to produce S2

2−, respectively, accompanyingwith nonradiative recombination. SPRM images the time trace of local RI around each single CdS NP and background region simultaneously during theillumination of blue light, from which the photocatalytic activity of single CdS NP is revealed from the large reduction in the local RI of surrounding mediumcaused by the hydrogen production. (Scale bar: 10 μm.) (B, Top) RI trajectories of NP1, NP2, and a background region during a continuous illumination of 2 hare displayed as red, blue, and black curves, respectively. (Bottom) Zoom-in trajectory of NP2 during 3,460th and 3,750th reveals alternatively switchingRI between a high value (photocatalysis OFF state) and a low value (photocatalysis ON state). (C) Statistical analysis reveals a truncated and full-range power-law distribution of ON and OFF state duration time, respectively. The ON and OFF probability densities of NP1 (red curves) are offset up by two scale factorsfor clarity.

Fang et al. PNAS | October 3, 2017 | vol. 114 | no. 40 | 10567

CHEM

ISTR

Y

Dow

nloa

ded

by g

uest

on

July

25,

202

0

nanoparticle, which was monitored continuously with SPRM. Thechange of the local RI over time was determined from the change inthe average SPRM intensity of the tail region for each nanoparticle.Surprisingly, we found that the photocatalytic production of

hydrogen at each of seven CdS nanoparticles switched on and offstochastically even though the light illumination was constant.Two representative time traces showing this intermittent natureof photocatalytic reactions are plotted in Fig. 1B (red and bluecurves) and more examples are given in SI Appendix, Fig. S6.Zoom-in of the time trace for one CdS nanoparticle (NP2 from3,460th to 3,750th s) reveals that the photocatalytic reaction wasswitched on and stayed ON for a certain time interval, and thenswitched off and stayed OFF for another time interval (see alsoMovie S1). As a control, the time trace of the local RI from aregion where there was no CdS nanoparticle was also measured(black curve in Fig. 1B), which showed a flat line with no char-acteristic switching events. Note that the time traces for differentCdS nanoparticles are uncorrelated, so ensemble average of alarge number of nanoparticles would have washed out such in-termittency. This observation underscores the need of single-nanoparticle studies to observe the intrinsic kinetics of thephotocatalytic reactions. Similar intermittency was also observedon CdS nanoparticles with different crystal structures and shapes(nanorods), indicating that the intermittent nature of photo-catalytic activity was a general phenomenon for various CdSnanomaterials regardless of their crystal structure and mor-phology (SI Appendix, Fig. S7). Please note that ∼80% of over300 individual CdS nanoparticles exhibited similar intermittentbehaviors in tens of independent experiments within a typicalrecording time of 3 h.The time durations of ON and OFF states of the photo-

catalytic reactions are marked by red and black dots in Fig. 1B(Bottom), respectively, showing stochastic distributions over timescales spanning from 0.16 s (the temporal resolution in thepresent work) to hundreds of seconds. Statistical analysis (26) ofthe individual switching events for different nanoparticles showsthat the OFF state follows a full-scale power-law distribution,and the ON state can be fitted with a truncated power-law dis-tribution (solid curves in Fig. 1C and SI Appendix, Figs. S6 andS8). The two distributions can be expressed as

p�toff

�=C · toff−αoff

pðtonÞ=�C ·

�kcutoff ton

�−αon for  ton ≥ 1=kcutoffC · exp

�−αon

�kcutoff ton − 1

��for  ton < 1=kcutoff

,

respectively, where p(ton) is the probability density of a certainon-event duration time (ton), C is a constant, αon is the power-lawcoefficient of ON events, and kcutoff is the truncation factor. Notethat the above functions are widely used to analyze fluorescencephotoblinking (26) and many other stochastic processes in phys-ical (27) and social sciences (28). A truncated power-law distri-bution indicates any process faster than kcutoff is prohibited whileprocesses slower than kcutoff are stochastically distributed (27).Control experiments were performed to show that the mea-

sured switching of the local RI was indeed due to the in-termittent hydrogen production. First, the local RI did notchange when the nanoparticles were not illuminated with light,or illuminated with red light (λ > 600 nm, SI Appendix, Fig. S9).Second, when replacing S2−/SO3

2− with HPO42− in the solution,

no switching in the local RI was detected (SI Appendix, Fig. S9),which was expected because S2−/SO3

2− was known to facilitatephotocatalytic hydrogen production (3, 29). These results alsodemonstrated that the photothermal effect was not responsiblefor the intermittent photocatalytic activity because it would beindependent of electrolytes. The illumination of blue light wasfound to increase the local temperature around single nano-particles by 0.1 °C as determined by a photothermal conversionapproach (30) (SI Appendix, Fig. S10), suggesting a small thermaldisturbance to the local reaction. Third, the position of individualCdS nanoparticles did not move at all over the entire experi-ment (hundreds of switching events in 2 h), indicating Brownianmotion of the nanoparticle was negligible. Fourth, CdS nano-particles exhibited excellent photostability in the presence ofS2−/SO3

2− without the sign of photocorrosion (SI Appendix, Fig. S11).Besides, an irreversible photocorrosion process was anticipatedto display a monotonic change in the SPR intensity (SI Appendix,Fig. S9) rather than switching between two states. Finally, the RIchanges due to interfacial processes, such as the adsorption and

0 60 120 180 240

-20

-10

0

SP

R In

tens

ity (m

Deg

)

Time (s)

0 10 200

10

20

No.

of E

vent

s

Time (s)-200 -250 -300 -350 -400 -450 -500

-25

0

SP

R In

tens

ity (m

Deg

)

Potential (mV)

Scan rate 2mV/s-300 -350 -400 -450 -5000

5

10

15

20

Bub

ble

liftim

e (s

)

Potential (mV)

Experimentalresult

Nano-sizedbubble

-5.0 -2.5 0.0 2.5 5.0-1.0

-0.5

0.0

0.5

Rel

ativ

e In

tens

ityHorizontal distance / µm

HomogeneousdiffusionA B

C D

Fig. 2. (A) Experimental SPRM image during hydrogen production (Center) reveals that the local RI reduction is confined in a volume smaller than theoptical diffraction limit. This observation supports the nanobubble hypothesis by comparing with the simulated SPRM image for nanobubble (Left)or homogeneous diffusion hypothesis (Right). (Scale bar: 1 μm.) (Lower) Horizontal profiles of three images crossing the center are displayed (red:nanobubble, blue: experimental, black: supersaturation). (B) RI trajectory of a single CdS NP during electrochemical reduced H2 evolution at a potentialof −400 mV vs. Ag/Ag2S. Each cycle, consisting of a gradual RI decrease and a rapid RI recovery, represents the growth and eclipse of a nanosized H2 bubble.(C) The bubble lifetime follows Gaussian distribution rather than a power-law distribution. (D) The bubble lifetime becomes shorter when electrodepotential scans toward more negative, indicating a faster H2 evolution rate at higher overpotential. The lifetime vs. potential is fitted with a single-rateexponential curve.

10568 | www.pnas.org/cgi/doi/10.1073/pnas.1708617114 Fang et al.

Dow

nloa

ded

by g

uest

on

July

25,

202

0

desorption of ions and small molecules, were too small to bedetected by SPRM.We have discussed that photocatalytic production of hydrogen

at a CdS nanoparticle leads to a decrease in the local RI.However, the produced hydrogen may either dissolve in the so-lution, or form nanosized hydrogen bubbles at the nanoparticle.If the former is true, we expect that the dissolved hydrogenmolecules diffuse away to the surrounding environment with arather large diffusion distance (200 μm) over 10 s (typical du-ration of the ON state) because of the large diffusion coefficientof hydrogen molecules in water (4.5 × 10−9 m2/s). We measuredthe spatial profile of RI drop and found that it was confinedwithin a region defined by the diffraction limit, much smallerthan the diffusion distance (Fig. 2A). This result indicated theformation of hydrogen nanobubbles with size below the diffrac-tion limit. We further confirmed this conclusion with a three-dimensional modeling (with COMSOL; SI Appendix, Fig. S13).The model simulation demonstrated that the generation of a66-nm nanobubble attaching to the CdS nanoparticle could pro-duce an SPRM image consistent with the experimental observation

(SI Appendix, Fig. S14). The present setup allows for the detectionof a nanobubble as small as 30 nm. From the slope of the accu-mulative profile of SPRM intensity trajectories of all seven CdSnanoparticles, one can estimate the single-nanoparticle reactionrate to be 1.1∼7.0 × 104 H2 molecules per second per nanoparticle(Fig. 3A), corresponding to a single-nanoparticle apparent quan-tum yield of 0.01∼0.06% (SI Appendix, Fig. S14). This value is closeto the reaction rate (7.4 × 104 H2 molecules per second pernanoparticle) we determined for ensemble catalysts with tradi-tional techniques (SI Appendix, Fig. S16). Most importantly, it wasfound that an induction time of 800∼1,900 s was required to ob-serve the decrease in SPR intensity after the blue-light illumination(Fig. 3B). During the induction time, the local reaction solutionhad to be saturated by photogenerated H2 molecules to facilitatethe formation of H2 bubbles. Presaturation of the reaction solutionwith H2, or increasing the surface density of CdS nanoparticles, wasable to significantly shorten the induction time (Fig. 3C), stronglysupporting that H2 nanobubbles were responsible for the decreasein SPR intensity. The generation of H2 molecules was furthersupported by analyzing the photochemical reaction products withgas chromatography (SI Appendix, Fig. S12).We believe that the photocatalytic reaction led to the gener-

ation of hydrogen nanobubbles with sizes balanced by productionof hydrogen at the nanoparticle and dissolution of hydrogen intothe surrounding solution. Once the photocatalytic activity stop-ped, the nanobubble disappeared rapidly due to the high Laplacepressure (31, 32), i.e., the hydrogen molecules dissolved in thesolution and diffused away from the nanoparticles, leading to therecovery of the local RI. The stop of photocatalytic activity couldnot be attributed to the surface blockade by nanobubble. If theblockade hypothesis is true, the nanobubbles would immediatelyregrow once they were collapsed. That is because the collapse ofnanobubble restarts the photocatalytic activity and leads to theimmediate regrowth of nanobubble if the hydrogen productionrate is constant. This is opposite to the experimental resultswhere long OFF events (no nanobubble) lasting up to hundredsto thousands of seconds were often observed.Despite the nanobubble formation, we show below that the

observed ON and OFF switching events reflected intrinsic in-termittent nature of the photocatalytic activity, rather than theformation and dissolution kinetics of the nanobubbles, by com-paring results between photocatalytic and electrochemical hy-drogen reduction. The electrochemical measurement wasperformed under the same condition as the photocatalytic ex-periment except that an applied electrochemical potential, in-stead of blue light, was used to trigger hydrogen production(SI Appendix, Fig. S15). An electrochemical reduction potentialof −400 mV vs. Ag/Ag2S also leads to the formation of nano-bubbles at the locations of individual CdS nanoparticles (Fig. 2B),but the distribution of bubble lifetimes follows a Gaussian dis-tribution (Fig. 2C), which is in contrast to the power-law distri-bution for photocatalysis (Fig. 1B). For fair comparison, thepotential was selected such that the bubble growth rate (nano-bubble lifetime ranging from 5 to 40 s) in the electrochemical re-action was similar to and sometimes much slower than that in thephotocatalytic reaction (lifetime ranging from 1 to 10 s). Anothersharp difference between the electrochemical reduction and pho-tocatalytic reactions is that the OFF state in the former does nothave a finite duration (Fig. 2B), even with lower reaction rate(slower kinetics; SI Appendix, Fig. S15B). In other words, once ananobubble disappears, a new nanobubble starts to form immedi-ately, indicating the electrochemical production of hydrogen neverstops. In contrast, in the case of photocatalytic production of hy-drogen, it stochastically stays off for a time interval ranging fromsubsecond to thousands of seconds before it switches on again.Increasing the overpotential simply reduces the average bubblelifetime but does not change the Gaussian distribution and the lackof time duration for OFF states, as shown in Fig. 2D and Movie S2.

0 2000 4000 6000 80000

2

4

6

NP 6

NP 5

NP 2

NP 1

NP 4

NP 3

Acc

umul

ated

am

ount

of H

2 Pro

duct

ion

at s

ingl

e N

Ps /

108 m

olec

ules

illumination time / s

NP 1 NP 2 NP 3 NP 4 NP 5 NP 6 NP 7 averaged

NP 7

0

500

1000

1500

2000

2500Presaturated with N2

High density CdS NPs presaturated with N2

Indu

ctio

n tim

e (s

)

Presaturated with H2

0 500 1000 1500 2000 2500-80

-60

-40

-20

0

20

SPR

Inte

nsity

(mD

eg)

Time (s)

Presaturated with N2

Presaturated with H2

Apply blue light illumina�onat this moment (�me zero).

Induc�on �me

Induc�on �me

A

B

C

Fig. 3. (A) Time traces of the accumulated amount of hydrogen productionfrom each single CdS NP and the averaged hydrogen production from theseseven NPs. (B) Representative SPR intensity curves for a single CdS NP in thereaction solution presaturated with N2 (black curve) and H2 (red curve), re-spectively. (C) The induction time was significantly reduced when presaturatingthe reaction solution with H2 (Left Bar), or when increasing the surface densityof CdS NPs (Right Bar). Over 30 NPs were analyzed under each condition.

Fang et al. PNAS | October 3, 2017 | vol. 114 | no. 40 | 10569

CHEM

ISTR

Y

Dow

nloa

ded

by g

uest

on

July

25,

202

0

The exponential dependence of the ON-state duration time withthe overpotential (Fig. 2D, Inset) demonstrates that the bubblelifetime is regulated by hydrogen production rate.We further studied the correlation between kcutoff and the

bulk hydrogen production rate by varying the concentration ofS2− ([S2−]) and the power density of blue-light illumination;both are confirmed to be relevant to the hydrogen productionrate (SI Appendix, Fig. S16) (3). When [S2−] was reduced from500 to 200 mM, intermittent photocatalysis remained but theON-state duration increased significantly (SI Appendix, Fig. S17).Statistical analysis of the ON-state duration reveals a sixfolddecrease in kcutoff associated with the reduction of [S2−] and asignificant shift of the truncation toward a longer time scale (Fig.4A). A similar analysis shows that kcutoff decreases with illuminationpower density (Fig. 4B). Plotting kcutoff vs. hydrogen production rateshows a linear relationship (Fig. 4C), suggesting that kcutoff reflectsthe photochemical reaction rate of single CdS nanoparticles. Wealso examined the effect of [S2−] and illumination power density onthe OFF-state duration distribution, and full-scale power-law dis-tributions remained in all cases (SI Appendix, Fig. S18).In addition to the ON/OFF event time, the growth rate of

nanobubble is also able to quantify the H2 generation rate asso-ciated with single nanoparticles (SI Appendix, Fig. S17B). If wecompare the growth rates of nanobubble in different ON eventsunder the same conditions, they are also stochastic and vary by upto 10× (SI Appendix, Fig. S17C). When increasing the apparentphotochemical reaction rate, growth rates of nanobubble also in-crease. We chose ON/OFF event time instead of growth rate inpower-law analysis, because the reliable curve fitting for growth

rate required significantly more data points and it tended to induceuncertainties when the signal/noise ratio was not good enough.A similar power-law distribution has been observed in fluores-

cence photoblinking of single quantum dot, which has been at-tributed to photoionization and multiple rates during carrierrecombination (8, 10). These processes may also contribute to theintermittent photocatalytic activity of single CdS nanoparticlesobserved here because of the intrinsic connection between thephotochemical reactions and the photophysical processes. For ex-ample, previous studies have suggested that the interfacial electrontransfer activity of semiconductor nanoparticles was modulated bythe fluorescence photoblinking dynamics (33). The absence ofpower-law distribution in electrochemical reduction of hydrogenfurther supports this point, as photophysical processes are solelyinvolved in the photocatalysis but not in the electrochemical re-duction. When looking at the photochemical reactions at singleCdS nanoparticles, multiple reaction sites could exist simulta-neously on the nanoparticle surface, resulting in a stochastic dis-tribution of the reaction energy barriers at each active site due tothe heterogeneous coordination states and dangling bonds (34, 35).Therefore, the photophysical processes as well as multiple in-terfacial electron transfer rates are believed to be the key for theobserved power-law distribution. While general power-law distri-butions of ON/OFF events are attributed to the intermittentphotophysical processes, the truncation (kcutoff) is a consequence ofthe interfacial photochemical reactions, because such truncation isa unique distribution that has been rarely observed in fluorescencephotoblinking. Therefore, it reflects the reaction rate at the single-nanoparticle level (Fig. 4C). The existence of multiple reactionsites was also supported by two experimental observations. First,multiple nanobubble generation sites were often found for CdSnanorods (SI Appendix, Fig. S7 E–H). Second, multiple statesrather than two states existed on the projection of SPR trajectorycurves on the intensity axes (SI Appendix, Fig. S20 and Fig. 5). Weattributed this feature to the different locations (vertical distancesto the substrate) and different sizes of the nanobubbles as a resultof the multiple reactive sites. SPRM signal of a bubble sensitivelyrelies on the vertical distance and the size. Note that the variedSPRM intensity does not affect the power-law analysis which relieson the duration time of ON/OFF events, rather than its intensity.To further support the above analysis, we simulated the

probability density of ON-state durations with a stochastic model

1 10 100 100010-6

10-5

10-4

10-3

10-2

10-1

200 mMNa2S

Prob

abili

ty d

ensi

ty (1

/s)

ON duration time (s)

500 mMNa2S

1 10 100 100010-6

10-5

10-4

10-3

10-2

10-1

Prob

abili

ty d

ensi

ty (1

/s)

On duration time (s)

14 mW mm-2 Blue Light 9 mW mm-2 Blue Light 6 mW mm-2 Blue Light

0 2 4 6 8 100.00

0.03

0.06

0.09

0.12

6 mW mm-2,500 mM Na

2S

14 mW mm-2, 200 mM Na2S

9 mW mm-2,500 mM Na

2S

k cut

-off

(s-1

)

H2 production rate (mL/hr)

14 mW mm-2,500 mM Na2S

1 10 100 100010-7

10-6

10-5

10-4

10-3

10-2

10-1

100

kcut-off

= 0.03 s-1

kcut-off = 0.3 s-1

kcut-off = 30 s-1

Prob

abili

ty d

ensi

ty (1

/s)

ON duration time (s)

A B C D

Fig. 4. (A) Probability density of ON events in the presence of 500 (magenta) and 500 (black) mM Na2S. (B) Probability density of ON events under the blue-light power density of 14 (blue), 9 (magenta), and 6 (black) mW·mm−2. (C) Correlation between kcutoff (extracted from the probability density curves in A andB) and bulk H2 production rate (measured from conventional photocatalysis experiment using ensemble materials). (D) Calculated truncated power-lawdistributions by adjusting the maximum rate kcutoff in a multiple-rate stochastic process.

-15 -10 -5 00

100

200

300

Cou

nts

SPR Intensity (mDeg)

NP1 (zoom in)

-15 -10 -5 00

100

200

300

Cou

nts

SPR Intensity (mDeg)

NP2 (zoom in)A B

Fig. 5. Zoom-in of the projection histograms of the corresponding SPR in-tensity of NP1 (A) and NP2 (B) reveal the existence of multiple states (mul-tiple ON and single OFF) rather than two states (single ON and single OFF).

10570 | www.pnas.org/cgi/doi/10.1073/pnas.1708617114 Fang et al.

Dow

nloa

ded

by g

uest

on

July

25,

202

0

(27, 28) involving multiple active sites, each with a reaction rate(SI Appendix, Fig. S19). If assuming 100 active sites with ratesuniformly distributed from 1 × 10−4 to 30 s−1, the calculatedprobability density displays a full-scale power-law distributionranging from 1,000 to 0.5 second (Fig. 4D, blue dots). In thiscase, kcutoff is 30 s−1. However, if reducing kcutoff to 0.3 and0.03 s−1, truncated power-law distributions are predicted by themodel (red and black dots), which reproduces key features of theexperiments (Fig. 4 A and B). The truncation point shifts towardlonger time scale for the smaller kcutoff, accompanying with lowerprobability densities in the plateau region and constant slopes inthe linear region. The agreement between the multiple reactionrates model and the experimental data supports the hypothesisthat multiple photochemical reaction rates stochastically distributedin a certain range with a maximal rate are responsible for the trun-cated power-law distribution of ON-state durations observed on singleCdS nanoparticles. The active sites at atomic scale are attributed tothe unsaturated Cd atoms with dangling bonds, as well as some bigcurvatures in the irregular CdS nanoparticles (contacts, edges, andcaves). Further efforts are certainly required to comprehensively un-derstand the structural basis of these active sites. Combining theSPRM technique proposed here and high-resolution electron mi-croscopy is a promising way to achieve this goal in the future.In summary, we report the intermittent photocatalytic activity

of single CdS nanoparticles, and propose a multiple reaction sitesmodel for the full-scale and truncated power-law distribution of ONand OFF events, respectively. As photoblinking is a general phe-nomenon for various types of semiconductor nanomaterials, thediscovery of the intermittent photocatalysis opens a field of semi-conductor-based photocatalysis and paves the way toward the ra-tional design and discovery of photocatalysts with excellentefficiency by studying the photocatalysis at single-nanocatalyst level.From the photophysical point of view, the present work also con-nects the single-nanoparticle photophysics and photochemistry,offering another opportunity to clarify the mechanistic origin of themysterious photoblinking of single-semiconductor nanomaterials.

Materials and MethodsExperimental Setups, Synthesis, and Characterizations of the CdS Nanoparticles.The plasmonic microscopy setup was built on an inverted total internal re-flection fluorescence microscope using gold-film-coated glass coverslip as thesensor chip. CdS nanoparticles (NPs) were synthesized according to the previouswork (29). The resulting CdS NPs were uncapped (or capped with S2−) to fa-cilitate the photochemical reactions. CdS NPs with the cubic-phase were syn-thesized following a similar protocol as mentioned above, by switching thevolume of Na2S and Cd(OAc)2. The CdS nanorods were synthesized with asolvent thermal method. Dynamic light scattering, diffuse reflection UV-visspectrum, X-ray powder diffraction, X-ray photoelectron spectroscopy, etc.were used to characterize CdS NPs. Please see SI Appendix for more details.All of the photochemical and electrochemical reactions were performed inaqueous solutions.

Power-Law Analysis. The power-law analysis of the SPRM trajectory curve wasachieved with a self-developed MATLAB code using a threshold-based al-gorithm, which has been routinely adopted to analyze the blinking fluo-rescence trajectory. The ON/OFF event was differentiated by the amount ofintensity fluctuation. A rapid decrease in the SPRM intensity (higher than auser-defined threshold) represents the beginning of an ON state, and a sharpincrease represents the end of anON state (and therefore the beginning of anOFF state). A detailed description on the MATLAB program is provided in SIAppendix, section 1.4. After collecting the duration times for a series of ON/OFF events, the probability density of each duration time was calculatedwith the following equation:

PðτiÞ=2× Nτ,i

Ntotal

ðτi+1 − τiÞ+ ðτi − τi−1Þ,

where P(τi) is the probability density of ON (OFF) event with duration time ofτi, Nτ,i is the number of ON (OFF) event with duration time of τi, Ntotal is thetotal number of ON (OFF) events. The unit of P(τi) is s

−1.

ACKNOWLEDGMENTS. We acknowledge financial support from the Na-tional Natural Science Foundation of China (Grants 21327902, 21527807,21522503, and 21605078), and the Natural Science Foundation of JiangsuProvince (BK20150013 and BK20150570).

1. Han Z, Qiu F, Eisenberg R, Holland PL, Krauss TD (2012) Robust photogeneration of H2 inwater using semiconductor nanocrystals and a nickel catalyst. Science 338:1321–1324.

2. Chen X, Liu L, Yu PY, Mao SS (2011) Increasing solar absorption for photocatalysis withblack hydrogenated titanium dioxide nanocrystals. Science 331:746–750.

3. Chen X, Shen S, Guo L, Mao SS (2010) Semiconductor-based photocatalytic hydrogengeneration. Chem Rev 110:6503–6570.

4. Roeffaers MBJ, et al. (2006) Spatially resolved observation of crystal-face-dependentcatalysis by single turnover counting. Nature 439:572–575.

5. Xu W, Kong JS, Yeh YTE, Chen P (2008) Single-molecule nanocatalysis reveals het-erogeneous reaction pathways and catalytic dynamics. Nat Mater 7:992–996.

6. Sambur JB, et al. (2016) Sub-particle reaction and photocurrent mapping to optimizecatalyst-modified photoanodes. Nature 530:77–80.

7. Nirmal M, et al. (1996) Fluorescence intermittency in single cadmium selenide nano-crystals. Nature 383:802–804.

8. Frantsuzov P, Kuno M, Janko B, Marcus RA (2008) Universal emission intermittency inquantum dots, nanorods and nanowires. Nat Phys 4:519–522.

9. Galland C, et al. (2011) Two types of luminescence blinking revealed by spectroelec-trochemistry of single quantum dots. Nature 479:203–207.

10. Cordones AA, Leone SR (2013) Mechanisms for charge trapping in single semiconductornanocrystals probed by fluorescence blinking. Chem Soc Rev 42:3209–3221.

11. Novo C, Funston AM, Mulvaney P (2008) Direct observation of chemical reactions onsingle gold nanocrystals using surface plasmon spectroscopy. Nat Nanotechnol 3:598–602.

12. Liu N, Tang ML, Hentschel M, Giessen H, Alivisatos AP (2011) Nanoantenna-enhancedgas sensing in a single tailored nanofocus. Nat Mater 10:631–636.

13. Seo D, Park G, Song H (2012) Plasmonic monitoring of catalytic hydrogen generationby a single nanoparticle probe. J Am Chem Soc 134:1221–1227.

14. Li L, et al. (2012) Metal oxide nanoparticle mediated enhanced Raman scattering andits use in direct monitoring of interfacial chemical reactions. Nano Lett 12:4242–4246.

15. Tachikawa T, Majima T (2010) Single-molecule, single-particle fluorescence imagingof TiO2-based photocatalytic reactions. Chem Soc Rev 39:4802–4819.

16. Zhou X, et al. (2012) Quantitative super-resolution imaging uncovers reactivity pat-terns on single nanocatalysts. Nat Nanotechnol 7:237–241.

17. Zhang Y, et al. (2015) Superresolution fluorescence mapping of single-nanoparticlecatalysts reveals spatiotemporal variations in surface reactivity. Proc Natl Acad Sci USA112:8959–8964.

18. Chen T, Zhang Y, Xu W (2016) Size-dependent catalytic kinetics and dynamics of Pdnanocubes: A single-particle study. Phys Chem Chem Phys 18:22494–22502.

19. Wang W, et al. (2011) Single cells and intracellular processes studied by a plasmonic-based electrochemical impedance microscopy. Nat Chem 3:249–255.

20. Wang W, et al. (2012) Label-free measuring and mapping of binding kinetics ofmembrane proteins in single living cells. Nat Chem 4:846–853.

21. Shan X, et al. (2012) Imaging the electrocatalytic activity of single nanoparticles. NatNanotechnol 7:668–672.

22. Labonté AP, Tripp SL, Reifenberger R, Wei A (2002) Scanning tunneling spectroscopyof insulating self-assembled monolayers on Au(111). J Phys Chem B 106:8721–8725.

23. Bain CD, et al. (1989) Formation of monolayer films by the spontaneous assembly oforganic thiols from solution onto gold. J Am Chem Soc 111:321–335.

24. Kolomenski A, Kolomenskii A, Noel J, Peng S, Schuessler H (2009) Propagation lengthof surface plasmons in a metal film with roughness. Appl Opt 48:5683–5691.

25. Halpern AR, Wood JB, Wang Y, Corn RM (2014) Single-nanoparticle near-infraredsurface plasmon resonance microscopy for real-time measurements of DNA hybrid-ization adsorption. ACS Nano 8:1022–1030.

26. Kuno M, Fromm DP, Hamann HF, Gallagher A, Nesbitt DJ (2001) “On”/”off” fluorescenceintermittency of single semiconductor quantum dots. J Chem Phys 115:1028–1040.

27. Clauset A, Shalizi CR, Newman MEJ (2009) Power-law distributions in empirical data.SIAM Rev 51:661–703.

28. Peterson GJ, Pressé S, Dill KA (2010) Nonuniversal power law scaling in the probabilitydistribution of scientific citations. Proc Natl Acad Sci USA 107:16023–16027.

29. Yan HJ, et al. (2009) Visible-light-driven hydrogen production with extremely highquantum efficiency on Pt-PdS/CdS photocatalyst. J Catal 266:165–168.

30. Chen Z, et al. (2015) Imaging local heating and thermal diffusion of nanomaterialswith plasmonic thermal microscopy. ACS Nano 9:11574–11581.

31. Craig VSJ (2011) Very small bubbles at surfaces-the nanobubble puzzle. Soft Matter 7:40–48.

32. German SR, et al. (2016) Electrochemistry of single nanobubbles. Estimating thecritical size of bubble-forming nuclei for gas-evolving electrode reactions. FaradayDiscuss 193:223–240.

33. Jin S, et al. (2010) Correlated single quantum dot blinking and interfacial electrontransfer dynamics. Chem Sci 1:519–526.

34. Huang L, et al. (2013) Effects of surface modification on photocatalytic activity of CdSnanocrystals studied by photoluminescence spectroscopy. Phys Chem Chem Phys 15:553–560.

35. Bryant GW, Jaskolski W (2005) Surface effects on capped and uncapped nanocrystals.J Phys Chem B 109:19650–19656.

Fang et al. PNAS | October 3, 2017 | vol. 114 | no. 40 | 10571

CHEM

ISTR

Y

Dow

nloa

ded

by g

uest

on

July

25,

202

0