ORIGINAL PAPER

Harmonic Mode-Locking www.lpr-journal.org

Revealing the Buildup Dynamics of Harmonic

Mode-Locking States in Ultrafast Lasers
Xueming Liu* and Meng Pang

1. Introduction
Harmonic mode-locking (HML) is an important technique enabling the
generation of high-repetition-rate ultrashort pulses. Using an emerging Ultrafast lasers with high repetition
rate have many potential applications in
time-stretch dispersive Fourier transform technique, the experimental multiple areas such as optical telecom-
observation of the entire buildup process of the passive HML state in an munications, frequency metrology,
ultrafast fiber laser is reported here. It is unveiled that the whole process of high-speed optical sampling, and data
HML buildup successively undergoes seven different ultrafast phases: raised storage.[1–9] Usually, fiber lasers have a
relaxation oscillation, spectral beating behavior, birth of a giant pulse, fundamental repetition rate of tens to
hundreds of MHz due to the difficulty
self-phase-modulation-induced instability, pulse splitting, repulsion and
of shortening the laser cavity.[10–15] To
separation of multiple pulses, and a stable HML state. It is observed that the increase the pulse repetition rate of fiber
multiple HML pulses originate from a single-pulse splitting phenomenon and lasers, a less technically challenging and
a remarkable breathing behavior occurs at an early stage of the HML buildup more convenient way is the harmonic
process. The numerical results confirm that the effects of dispersive wave, mode-locking (HML) scheme, where
gain depletion and recovery, and acoustic wave play key roles in the earlier, multiple pulses are evenly spaced in the
cavity.[16] Stable HML operations of fiber
middle, and later stages of this HML buildup process, respectively; as well, lasers have been successfully achieved
the acoustic resonance in the single-mode fiber stabilizes the final HML state by means of active mode-locking,[17,18]
of lasers. passive mode-locking,[19–21] and sub-
loops.[13,22–24] Among them, the passive
HML scheme has, from a practical
point of view, the intrinsic advantage of repetition-rate self-
stabilization. Passive HML fiber lasers, which are able to scale the
Prof. X. Liu pulse repetition rate to multiple GHz,[1,3,11] even up to 22 GHz
State Key Laboratory of Modern Optical Instrumentation (corresponding to the 928th HML),[25] have been reported re-
College of Optical Science and Engineering
cently.
Zhejiang University
Hangzhou 310027, China A fiber laser can produce multiple optical pulses within its
E-mail: liuxm@zju.edu.cn; liuxueming72@yahoo.com cavity when the pump power is increased to a higher level.
Prof. X. Liu These multiple pulses are generated by the noise-seeded mod-
Institute for Advanced Interdisciplinary Research ulation instability that leads to the unstable breakup and pulse
Nanjing University of Aeronautics and Astronautics splitting.[8,10,26,27] The attractive force between the adjacent pulses
Nanjing 210016, China
can lead to the formation of pulse groups such as stable bound
Prof. X. Liu
College of Astronautics states,[28] unstable bunches of pulses,[29] and even soliton rains.[30]
Nanjing University of Aeronautics and Astronautics In contrast, the repulsive force between them normally drives the
Nanjing 210016, China pulses to move away from each other, resulting in non-uniform
Prof. X. Liu or uniform pulse distribution throughout the cavity. The HML
School of Physics and Electronic Science generated in passively mode-locked lasers attributes to the long-
Hunan University of Science and Technology
range interaction of pulses by means of the acoustic-wave (AW)
Xiangtan 411100, China
effect,[16,31,32] the non-soliton component (i.e., the dispersive
Dr. M. Pang
State Key Laboratory of High Field Laser Physics wave (DW) or continuous wave) of radiation in the cavity,[33–35]
Shanghai Institute of Optics and Fine Mechanics and the gain depletion and recovery (GDR) mechanism.[36–38]
Chinese Academy of Sciences The passive HML phenomena have been demonstrated in a
Shanghai 201800, China wide variety of theoretical and experimental configurations
Dr. M. Pang in mode-locked lasers.[39–45] A 285th HML laser was reported
Max Planck Institute for the Science of Light
Straudtstrasse 2, 91058 Erlangen, Germany in 1993.[46] Usually, the HMLs are measured at the ultimately
stable state and some theoretical models are constructed on
The ORCID identification number(s) for the author(s) of this article the basis of the phenomenological hypothesis. Gordon and Gat
can be found under https://doi.org/10.1002/lpor.201800333 proposed a statistical mechanics model that the entropy is an
DOI: 10.1002/lpor.201800333 essential ingredient in the theory of mode locking.[47,48] So far,

Laser Photonics Rev. 2019, 13, 1800333 1800333 (1 of 9) 

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com
Figure 1. Schematic diagram of the experimental setup for the mode-
locked laser (on the left) and its ultrafast characterization line (on
the right). Timing data and real-time spectral data are achieved with
direct detection and by dispersing the pulses in 5-km dispersion-
compensating fiber (DCF) prior to detection, respectively. Inset: cross
section of the polarization-maintaining fiber. EDF, erbium-doped
fiber; LD, laser diode; CNT-SA, carbon nanotube saturable absorber;
WTI, hybrid combiner of wavelength division multiplexer, tap coupler
and isolator; OSA, optical spectrum analyzer; RF, radio frequency;
PD, photodetector.
the entire buildup process of passive HML has not been reported
in detail. The origin of HML buildup process is still unclear,
and no satisfactory theoretical description has been demon-
strated for unveiling the underlying mechanism of the HML
operation.
The spacing between multiple pulses operated in HML
undergoes a random change due to the presence of temperature
variations and mechanical vibrations. As a result, the HML op-
eration in the ordinary non-polarization-maintaining (non-PM)
fiber lasers can only remain stable in a short time-scale, and
even such operation is discontinued by these perturbations.
The controllable HML operation of passively mode-locked
fiber lasers, on the other hand, has been proved quite hard to
be achieved. The all-PM fiber lasers can be immune to these
variations and vibrations.[49] Coupling between cavity modes and
the acoustic resonance in a fiber provides strong and optome-
chanical interactions, resulting in “control elements” that enable
controllable and stable HML operation of lasers.[1] The emerging
time-stretch dispersive Fourier transform (TS-DFT) technique
can experimentally resolve the real-time evolution of ultrafast
optical phenomena.[50–55] Based on the TS-DFT technique to-
gether with the all-PM, specially designed, optomechanical laser
cavity, we have experimentally investigated the onset dynamics
of passive HML in detail for the first time to our knowledge. We
experimentally demonstrate that the buildup process of HML
includes such stages as the raised relaxation oscillation, beating
dynamics, birth of a giant pulse, self-phase modulation (SPM)-
induced instability, pulse splitting, repulsion and separation of
multiple pulses, and the stable HML state. It is revealed that
the HML operation of the laser originates from the splitting of
a single giant pulse, and the breathing pattern appears at the
early stage of the HML buildup process. Numerical simulations
corroborate that the DW, GDR, AW, and optomechanical inter-
action contribute to the buildup and stabilization of the passive
HML.
Laser Photonics Rev. 2019, 13, 1800333 1800333 (2 of 9)
www.lpr-journal.org
2. Experimental Setup
Recently, a special type of photonic crystal fiber (PCF) was used to
realize HML lasers, which increases the stabilizing action of the
AW-mediated force.[1,11] Here, we use commercial PM fibers to
construct the HML laser, leading to a simpler laser configuration.
The schematic diagram of the specially designed all-PM HML
laser used in the experiments is shown in Figure 1. The laser sys-
tem is composed of a polarization-insensitive carbon nanotube
saturable absorber (CNT-SA), a 1-m-long PM erbium-doped fiber
(EDF) with 110 dB m−1 absorption at 980 nm, a 20-m-long PM
single-mode fiber (SMF) (see the cross section in Figure 1), some
PM SMF pigtails, and a PM hybrid combiner of a wavelength
division multiplexer, a tap coupler, and an isolator (WTI). The
PM hybrid combiner (i.e., WTI) is used to ensure the unidirec-
tional operation, extract intracavity power with a ratio of 10%, and
launch the pump power from the laser diode (LD) into the gain
fiber. The CNT film here has a modulation depth of ࣈ12% and
a nonsaturable loss of ࣈ55%, serving as a mode-locker to initi-
ate the pulse operation. The group-velocity dispersions of SMF
(Panda PM1550-XP, Nufern) and EDF are estimated to be about
−22 and 30 ps2 km−1 at 1550 nm, respectively. The total cavity
length is ࣈ24.3 m, corresponding to a fundamental repetition
rate of ࣈ8.52522 MHz (i.e., 117.3 ns of the cavity round-trip [RT]
time). The real-time temporal and time-averaged spectral detec-
tions for pulses are recorded with a high-speed real-time oscillo-
scope and an optical spectrum analyzer (OSA), respectively. The
TS-DFT technique is implemented by temporally stretching the
pulses in a 5-km-long dispersion-compensating fiber (DCF) with
a dispersion of about −160 ps (nm·km)−1 .
3. Experimental Results
The pulse sequence circulating in the cavity can coherently drive
the acoustic vibration in the fiber, and simultaneously, the back-
action in the form of a refractive index modulation can affect
these pulses.[1,56] When the integer multiple frequency of the Nth
HML locates near the peak of optomechanical gain, the mechan-
ical resonance is excited, and then each pulse in the sequence is
trapped within one acoustic vibration cycle. As a result, N pulses
are evenly distributed inside the laser cavity. The experimental
setup for the passive HML laser is shown in Figure 1. This all-PM
mode-locked fiber laser can stably deliver the HML pulses from
the first to sixth order by gradually increasing the pump power.
The detailed theoretical explanation is shown in Section 4 below.
Figure 2a,b shows the experimental results recorded over the
entire buildup process of the fifth HML without and with the TS-
DFT technique, respectively. The recorded time series are seg-
mented with respect to 117.3 ns (i.e., the cavity RT time), and then
the buildup process of HML can be depicted by the RT time and
the RT number. The vertical and horizontal axes in Figure 2a,b
illustrate the time within a single RT and the dynamics across
consecutive RTs, respectively. Figure 2a,b demonstrates the en-
tire evolution process from noisy lasing to single-pulse mode-
locking, eventually to stable HML. Obviously, a big corner is ob-
served at ࣈ1.06 × 105 RTs, before and after which the evolution
trajectories of the intra-cavity pulses are quite different, indicat-
ing a quite obvious decrease of the pulse peak power at this cor-

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.lpr-journal.org

Figure 2. Entire buildup process of HML in a passively mode-locked laser. a,b) Experimental real-time characterizations of the fifth HML without (a) and
with (b) the TS-DFT technique, respectively. a,b) Left and middle: The whole buildup process from noisy lasing to single pulse mode-locking, finally to the
stable HML. The interval time from the last frame in the left plot to the first frame in the middle plot is about 5 min. a,b) Right: Exemplary single-shot data,
corresponding to the last frame in middle plot. The big corner is located at ࣈ1.06 × 105 round-trips (RTs). c) Conceptual representation of seven stages
during the buildup process of HML, successively undergoing the raised relaxation oscillation, beating dynamics, birth of single pulse, SPM-induced
instability, pulse splitting, repulsion and separation of multiple pulses, and pulse lock of HML via acoustic resonance. a.u. denotes arbitrary unit.

ner. It is shown that the single pulse after the corner has lower
peak power than before, leading to a faster velocity of the pulse
due to the relatively lower nonlinear refractive index. Figure 2c
exhibits the conceptual representation of different stages during
the entire buildup process of the HML state, including several
nonlinear phenomena and ultrafast processes.
The shot-by-shot spectral information of HML can be captured
by means of the TS-DFT technique, as shown in Figure 2b. To
reveal the details of spectral evolution of HML buildup, the ex-
panded view and 3D picture at the beginning stage of Figure 2b
are redrawn in Figures 3 and 4a, respectively. Figure 3 demon-
strates the nascent dynamics of HML buildup measured via the
TS-DFT technique, where Figures 3a and 3b illustrate the 3D pic-
ture and the corresponding top-view graph at the early stage of
passive HML buildup, respectively. The onset of passive HML
includes the raised relaxation oscillation and beating dynamics.
Two laser spikes occur at ࣈ1.71 × 104 and ࣈ1.88 × 104 RTs dur-
ing the raised relaxation oscillation stage. Note that, in the raised
relaxation oscillation, the second laser spike is stronger than the
first one, contrary to the damped relaxation oscillation where the
second one is weaker than the first one.[15,57] Each laser spike con-
tains a large number of sub-nanosecond pulses, as shown in Fig-
ure 3a. The obvious beating behavior is observed from ࣈ1.9 × 104
to 2.08 × 104 RTs. The interference pattern can be seen clearly in
Figure 3c, which is the enlargement of the A region in Figure 3b.
Note that only spontaneous emission noise occurs before ࣈ1.71
× 104 RTs.
Figure 4a demonstrates that, after the raised relaxation oscil-
lation and beating dynamics, the buildup process of HML ex-
periences different stages such as the single-pulse birth, SPM-
induced instability, pulse splitting, and repulsion and separation
of multiple pulses. Note that the spectral profiles are modulated
and cleaned before and after the big corner, respectively. In the
spectral domain, therefore, we can see the transition from the
modulation instability dynamics to a regime in which multiple
pulses are generated. In fact, the theoretical and experimental re-
sults show that the modulation instability is a fundamental pro-
cess of nonlinear science, leading to pulse splitting which gives
birth to multiple pulses.[8,10,26,27] After pulse splitting, the spectra
become smooth, as shown in Figure 4a. Figure 4b,c illustrates the
temporal and spectral profiles at 2.15 × 104 and 1.01 × 105 RTs,
respectively, corresponding to the operations of single pulse and
SPM-induced instability. The time-averaged spectral profile mea-
sured by the OSA is shown as the red curve in Figure 4c. The last
frame in the left plot of Figure 2a (corresponding to 8.43 × 105
RTs) is shown in Figure 4b, where five pulses coexist in the laser

Laser Photonics Rev. 2019, 13, 1800333 1800333 (3 of 9) 

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.lpr-journal.org

Figure 3. Early stage of HML buildup captured by using the TS-DFT technique, corresponding to the RT numbers from 1.5 × 104 to 2.2 × 104 . Note
that only noise background occurs before 1.5 × 104 RTs. a) The intensity profile (z-axis) evolves along with the intra-cavity time (x-axis) and the RT
number (y-axis). The onset of passive HML includes the raised relaxation oscillation and beating dynamics. There are two laser spikes during the raised
relaxation oscillation and each laser spike includes lots of pulses. b) The top-view graph of (a). c) Close-up of the data from the A region of (b), revealing
the interference pattern for the beating dynamics.

cavity. The experimental results show that the pulse energy of
each pulse at 8.43 × 105 RTs (see red curve in Figure 4b) is about
a fifth of pulse energy at 2.15 × 104 RTs (see blue curve in Fig-
ure 4b). It is obviously revealed that the multiple pulses of HML
are split from a single giant pulse.
Figure 5a illustrates the Fourier transform of each single-shot
spectrum in Figure 2b, corresponding to a field autocorrelation
of the momentary pulses, which traces the evolution of temporal
separation between pulses.[15,51] Figures 5b and 5c are enlarge-
ments of the A and B regions in Figure 5a, respectively. Fig-
ure 5b reveals that only single pulse exists in the cavity from
ࣈ0.2 × 105 to 0.3 × 105 RTs (an example is shown as blue curves
in Figure 4b,c), and in succession, weaker modulation instabil-
ity is induced by SPM.[58] The SPM-induced instability gradually
increases until 1.06 × 105 RTs (an example is shown as black
curves in Figure 4b,c), finally leading to the pulse splitting which
gives birth to five pulses, as shown in Figure 5c. Figure 5c re-
veals that multiple pulses are generated during the process from

Laser Photonics Rev. 2019, 13, 1800333 1800333 (4 of 9) 

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.lpr-journal.org

Figure 4. HML buildup stage after the raised relaxation oscillation and beating dynamics. a) The buildup process of HML from 1.835 × 104 to 1.23 × 105
RTs, which is the close-up of local region in Figure 2b. b) Temporal profiles at 2.15 × 104 , 1.01 × 105 , and 8.43 × 105 RTs (corresponding to the last
frame in left plot of Figure 2a), respectively. c) Spectral profiles at 2.15 × 104 and 1.01 × 105 RTs, respectively, corresponding to the cross sections of
(a). The red curve in (c) is measured by the time-averaged OSA and the other two curves are captured by a high-speed real-time oscilloscope.

ࣈ1.06 × 105 to 1.13 × 105 RTs, corresponding to the region near
the big corner (see Figures 2 and 4). The SPM-induced instabil-
ity, recorded from ࣈ0.3 × 105 to 1.06 × 105 RTs, is the crucial
stage for generating multiple intra-cavity pulses.[10,58] Figure 5a
demonstrates that the evolution trajectory of multiple pulses at
the region beyond 1.13 × 105 RTs is similar to that of single pulse
at the region from ࣈ0.2 × 105 to 0.3 × 105 RTs.
To uncover the underlying mechanism of pulse evolution in
the HML buildup, the close-up of local region in Figure 2a is re-
drawn to Figure 6. The 3D picture, demonstrating the repulsion
and separation among the five pulses (see Figure 6a), reveals that
the pulses are separated one by one. For instance, the first, sec-
ond, third, and fourth pulses are separated from the pulse group
at about 1.5 × 105 , 2.5 × 105 , 3.8 × 105 , and 4.8 × 105 RTs, re-
spectively. Note that the breathing dynamics with the period of
ࣈ2.1 × 104 RTs is experimentally observed in this stage, similar
to the previous report.[54] Figure 6b, which is redrawn as the 3D
picture from the data of the middle plot of Figure 2b, displays the
evolution of stable fifth HML after about 5 min of the pump start-
up. The five pulses are equidistant with the spacing of 23.46 ns
and they can stably propagate across the laser cavity without the
breathing behavior. The experimental observations show that the
successive switch-on/off can yield quite similar evolution pro-
cesses to the same stable HML state with seven typical stages as
shown in Figure 2c, while the detailed evolution trajectories could
be different from time to time. (See one more recorded results as
Figure S1, Supporting Information.) As a result, the breathing
pattern occurs at the early stage of HML buildup (see Figure 6a
and Figure S1, Supporting Information), while no breathing pat-
tern appears at the stable HML state (note that, here, the sepa-
ration between pulses is quite large by comparing to the pulse
duration) (see Figure 6b).
Although only the fifth HML is reported here in detail, the en-
tire process of HML buildup, which experiences seven different
ultrafast stages (see Figure 2), is independent of the order num-
ber of the final HML state. The experimental results show that the
entire buildup processes of the second to sixth HMLs are similar
to the results shown in Figure 2. Two examples (i.e., the buildup
processes of the second and sixth HMLs) are shown in Figure S2,
Supporting Information. To further reveal the buildup process of
HML, we record the videos of the second to sixth HMLs captured
by radio-frequency spectrum analyzer (see Videos S1 to S5, Sup-
porting Information, for the full animation). These videos exhibit
that the buildup time of HML is about a minute due to the long-
range nature of the weak forces. The splitting processes of pulses
in these HMLs can be clearly observed through a real-time oscil-
loscope, as shown in the recorded videos (see Videos S6 and S7,
Supporting Information, for the full animation). The proposed
all-PM ultrafast laser can operate at the HML state with long-term
stability (e.g., the spectral profile and pulse duration are hardly

Laser Photonics Rev. 2019, 13, 1800333 1800333 (5 of 9) 

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.lpr-journal.org

Figure 5. a) Field autocorrelation of each single-shot spectrum in Figure 2b, tracing the separation between pulses. The inset is the local enlargement
of SPM-induced stability. b,c) Close-ups of the data from A and B regions of (a), respectively, revealing some details. The single-pulse operation, SPM-
induced instability, pulse splitting, and multi-pulse operation occur at the regions from ࣈ0.2 × 105 to 0.3 × 105 , ࣈ0.3 × 105 to 1.06 × 105 , ࣈ1.06 × 105
to 1.13 × 105 , and >1.13 × 105 RTs, respectively.

changeable for >100 h), so that it provides a platform to measure
the real-time buildup process of passive mode-locking from the
fundamental to higher harmonics.
The buildup process of mode-locked lasers is quite complex
and diverse. For instance, the single-pulse buildup undergoes the
evolution from continuous wave[10,59] or relaxation oscillation[15]
or transient bound-state[60] to stable pulse operation. Here, we re-
veal that the HML lasers originate from the pulse fragmentation
process, in which a giant pulse is first formed, then fragmented
into smaller identical pulses, and evenly distributed in the cavity.
Other ways may exist in the buildup process of HML lasers, es-
pecially when the number of pulses is high (e.g., N >10 pulses).
Besides the SPM-induced instability, dissipative processes are
also instrumental in destabilizing the giant pulse because dis-
sipative solitons can contain elements of chaotic dynamics sim-
ilar to that of a strange attractor in low-dimensional systems.[61]
Multiple theoretical models have been proposed to simulate the
starting of mode-locked lasers, for example, a statistical mechan-
ics model based on the idea of the system crossing an entropic
barrier[47,48] and a two-step method based on the rate equations
and the roundtrip circulating-pulse method.[60]
4. Stability Analysis and Theoretical Confirmation
When the pulse-to-pulse spacing is hundreds of times longer
than the pulse duration, the coherent overlap between adjacent
pulses is so weak that their interaction can be ignored. The long-
range interactions through DW, GDR, and AW then play cru-
cial roles in the repulsion and separation of multiple intra-cavity
pulses. The numerical simulations show that the DW, GDR, and
AW dominate the repulsive force when the pulse-to-pulse spac-
ing is less than ࣈ0.4 ns, from ࣈ0.4 to ࣈ23 ns, and more than
ࣈ23 ns, respectively (see Figures S4a, S6, and S11, Supporting
Information).
The SMF considers only the LP01 optics mode but lots of acous-
tic modes.[1,62] The first twenty acoustic modes and their gain
factors are shown in Figure S9a and Table S1, Supporting Infor-
mation. Both theoretical and experimental results show that the
acoustic vibration can be significantly enhanced by a sequence of
pulses with a repetition rate lying within the optomechanical gain
band and the acoustic loss is balanced by the gain from the elec-
trostrictive driving forces of the pulse sequence.[1] The change of
refractive index modulation via the optically driven acoustic vi-
bration tends to the maximum value when the pulse sequence
with an integer multiple (e.g., M) repetition rate can match the
maximum value of the optomechanical gain spectrum. As a re-
sult, the enhanced optoacoustic effect enables multiple pulses to
interact intensely (similar to the formation of stably optomechan-
ical bound states[1] ), and then the automatic and robust locking
between M repetition rate and acoustic frequency provides the
pulse sequence immune to environmental perturbations. An ex-
ample is exhibited in Figure 7a where the pulse repetition rate
of the fifth HML closely matches the resonant frequency of the
fourth (R04 ) acoustic mode in the SMF, giving the maximum
acoustic gain Gmax of ࣈ5.5 × 10−3 m−1 W−1 . In this case, the shift

Laser Photonics Rev. 2019, 13, 1800333 1800333 (6 of 9) 

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com www.lpr-journal.org

Figure 6. Experimental real-time measurement of the fifth HML buildup with 3D pictures. a) Repulsion and separation in the beginning stage of HML
buildup, in which data are from the left plot of Figure 2a. The buildup process of HML from 1.07 × 105 to 8.43 × 105 RTs. The breathing pattern occurs
with a period of ࣈ2.1 × 104 RTs. b) Stable stage of HML, in which data are from the middle plot of Figure 2b. Note that no breathing pattern is observed
in the stable stage.

between the acoustic resonant frequency of R04 mode and the 20-
fold fundamental cavity frequency (corresponding to the fourfold
[i.e., M = 4] repetition rate of the fifth HML) is ࣈ0.07 MHz, as
shown in the inset of Figure 7a. For the second and fourth HMLs,
M is 10 and 5, respectively. However, the third and sixth HMLs
most closely match the seventh (R07 ) acoustic mode, correspond-
ing to Gmax of ࣈ4.6 × 10−3 m−1 W−1 and the frequency shift of
ࣈ1.14 MHz (see Figure S9c, Supporting Information). The exper-
imental results demonstrate that our all-PM laser cannot stably
deliver some higher-order HMLs, for example, the seventh and

Figure 7. Contributions of acoustic resonance and optomechanical effect to the locking and stabilization of HML. a) Acoustic gain spectrum (red curve)
for the fourth (R04 ) acoustic mode in SMF. The integer multiple repetition rate of the second, fourth, and fifth HML is ࣈ170.5 MHz, corresponding to
an ࣈ0.07 MHz shift from the gain peak frequency at 170.57 MHz. The inset shows the close-up at the acoustic gain peak region. b) Relative velocity (red
curve) induced by the acoustic vibration of the R04 mode, which locks the pulse in the equilibrium position. Inset shows the trapping potential due to
the “spring” effect.

Laser Photonics Rev. 2019, 13, 1800333 1800333 (7 of 9) 

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com
eighth HMLs. The numerical simulations and the detailed the-
oretical explanations are shown in Figure S9d,e and Section S6,
Supporting Information.
The relative velocity induced by the acoustic resonance can
be up to ࣈ1.1 × 10−2 m s−1 and the pulse can be locked at the
equilibrium position, as shown in Figure 7b, where the blue and
black arrows represent the directions of AW-induced force. The
detailed depiction is shown in Figure S10, Supporting Informa-
tion. The proposed laser cavity has a net anomalous dispersion,
and the repetition rate of the driving pulse sequence lies on the
low-frequency side of the optomechanical gain band, resulting in
a “trapping potential” for the pulse timing, as shown in the inset
of Figure 7b. The second-order optomechanical term cooperates
with the average cavity dispersion to generate a “spring” effect
on the pulse timing jitter (t0 ). The theoretical explanation for
retiming mechanism is shown in ref. [1] in detail.
The timing jitter of pulse trains here is at several ps level, sim-
ilar to that of conventional, passive HML fiber lasers.[16] This is
because the underlying mechanism of all these lasers is based
on the weak optoacoustic effect in SMFs. From a practical point
of view, a feedback control system can significantly reduce the
timing jitter,[19] even to tens of fs.[63,64]
5. Conclusions
By means of the TS-DFT technique and a skillfully designed
all-PM laser cavity (it can induce the acoustic resonance), we have
experimentally tracked the formation and evolution of passive
HML in an ultrafast laser, unveiling its entire buildup process for
the first time to our best knowledge. Our experimental results
reveal that the HML buildup successively experiences seven
different stages, that is, raised relaxation oscillation, beating
dynamics, single-pulse birth, SPM-induced instability, pulse
splitting, repulsion and separation of multiple pulses, and stable
HML due to the optoacoustic effect. We discover that the stable
HML originates from a giant pulse, which can split into several
intra-cavity pulses due to the SPM-induced instability, featuring
the breathing dynamics at the early stage of the HML buildup
process. The numerical simulations confirm that the DW, GDR,
and AW dominate the repulsive force among adjacent pulses on
the earlier, middle, and later stages of HML buildup process,
respectively. By means of the optoacoustic effect that induces
a trapping potential, the acoustic resonance can stabilize the
mode-locking of laser at different harmonics (from the first to
sixth order at the appropriate pumping strength) with perfect
long-term stability. The comprehensive studies of these ultrafast
processes can both broaden our horizons in complex nonlinear
systems and contribute to laser design and applications (espe-
cially in the high repetition-rate operation and its steady-state
performance). These results could also provide some new per-
spectives in the evolution and dynamics of ultrafast phenomena
and bring useful insights into nonlinear science and applications.
Supporting Information
Supporting Information is available from the Wiley Online Library or from
the author.
Laser Photonics Rev. 2019, 13, 1800333 1800333 (8 of 9)
www.lpr-journal.org
Acknowledgements
The authors thank X. Yao, L. Hang, and Y. Zhang for fruitful discussions.
The work was partially supported by the National Natural Science Foun-
dation of China under Grant Agreements 61525505, 11774310, and by
the Key Scientific and Technological Innovation Team Project in Shaanxi
Province (2015KCT-06).
Conflict of Interest
The authors declare no conflict of interest.
Keywords
harmonic, mode-locked laser, real-time spectroscopy, transient
phenomena
Received: December 2, 2018
Revised: May 23, 2019
Published online: August 1, 2019
[1] M. Pang, W. He, X. Jiang, P. St. J. Russell, Nat. Photonics 2016, 10,
454.
[2] P. Grelu, N. Akhmediev, Nat. Photonics 2012, 6, 84.
[3] N. Tarasov, A. M. Perego, D. V. Churkin, K. Staliunas, S. K. Turitsyn,
Nat. Commun. 2016, 7, 12441.
[4] M. E. Fermann, I. Hartl, Nat. Photonics 2013, 7, 868.
[5] T. Udem, R. Holzwarth, T. W. Hansch, Nature 2002, 416, 233.
[6] U. Keller, Nature 2003, 424, 831.
[7] K. Krupa, K. Nithyanandan, U. Andral, P. Tchofo-Dinda, P. Grelu, Phys.
Rev. Lett. 2017, 118, 243901.
[8] J. M. Dudley, F. Dias, M. Erkintalo, G. Genty, Nat. Photonics 2014, 8,
755.
[9] L. G. Wright, D. N. Christodoulides, F. W. Wise, Science 2017, 358, 94.
[10] J. Peng, M. Sorokina, S. Sugavanam, N. Tarasov, D. V. Churkin, S. K.
Turitsyn, H. Zeng, Commun. Phys. 2018, 1, 20.
[11] M. Pang, X. Jiang, W. He, G. K. L. Wong, G. Onishchukov, N. Y. Joly,
G. Ahmed, C. R. Menyuk, P. St. J. Russell, Optica 2015, 2, 339.
[12] B. C. Collings, K. Bergman, W. H. Knox, Opt. Lett. 1998, 23, 123.
[13] S. S. Jyu, L. G. Yang, C. Y. Wong, C. H. Yeh, C. W. Chow, H. K. Tsang,
Y. Lai, IEEE Photonics J. 2013, 5, 1502107.
[14] J. N. Kutz, SIAM Rev. 2006, 48, 629; X. Liu, Phys. Rev. A 2010, 81,
023811.
[15] X. Liu, X. Yao, Y. Cui, Phys. Rev. Lett. 2018, 121, 023905.
[16] A. B. Grudinin, S. Gray, J. Opt. Soc. Am. B 1997, 14, 144.
[17] Y. Wang, S. Y. Set, S. Yamashita, APL Photonics. 2016, 1, 071303.
[18] Y. Lee, C. Tsai, Y. Chi, Y. Lin, G. Lin, IEEE J. Quantum Electron. 2015,
51, 1300111.
[19] Q. Hao, Y. Wang, P. Luo, H. Hu, H. Zeng, Opt. Lett. 2017, 42, 2330.
[20] M. E. Fermann, J. D. Minelly, Opt. Lett. 1996, 21, 970.
[21] A. Martinez, S. Yamashita, Opt. Express 2011, 19, 6155.
[22] E. Yoshida, Y. Kimura, M. Nakazawa, Appl. Phys. Lett. 1992, 60, 932.
[23] Y. Zhao, S. S. Min, H. C. Wang, S. Fleming, Opt. Express 2006, 14,
10475.
[24] M. L. Dennis, I. N. Duling, Electron. Lett. 1992, 28, 1894.
[25] C. Lecaplain, P. Grelu, Opt. Express 2013, 21, 10897.
[26] M. Närhi, B. Wetzel, C. Billet, S. Toenger, T. Sylvestre, J. Merolla, R.
Morandotti, F. Dias, G. Genty, J. M. Dudley. Nat. Commun. 2016, 7,
13675.
[27] D. R. Solli, G. Herink, B. Jalali, C. Ropers, Nat. Photonics 2012, 6, 463.

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
www.advancedsciencenews.com
[28] P. Grelu, J. M. Soto-Crespo, J. Opt. B: Quantum Semiclassical Opt.
2004, 6, S271.
[29] R. Gumenyuk, O. G. Okhotnikov, J. Opt. Soc. Am. B 2012, 29, 1.
[30] K. Sulimany, O. Lib, G. Masri, A. Klein, M. Fridman, P. Grelu, O. Gat,
H. Steinberg, Phys. Rev. Lett. 2018, 121, 133902; S. Chouli, P. Grelu,
Phys. Rev. A 2010, 81, 063829.
[31] A. B. Grudinin, D. J. Richardson, D. N. Payne, Electron. Lett. 1993, 29,
1860.
[32] S. Gray, A. B. Grudinin, W. H. Loh, D. N. Payne, Opt. Lett. 1995, 20,
189.
[33] W. H. Loh, A. B. Grudinin, V. V. Afanasjev, D. N. Payne, Opt. Lett.
1994, 19, 698.
[34] L. Socci, M. Romagnoli, J. Opt. Soc. Am. B 1999, 16, 12.
[35] J. S. Crespo, N. Akhmediev, P. Grelu, F. Belhache, Opt. Lett. 2003, 28,
1757.
[36] J. Nathan Kutz, B. C. Collings, K. Bergman, W. H. Knox, IEEE J. Quan-
tum Electron. 1998, 34, 1749.
[37] D. A. Korobko, O. G. Okhotnikov, I. O. Zolotovskii, Opt. Lett. 2015,
40, 2862.
[38] R. Weill, A. Bekker, V. Smulakovsky, B. Fischer, O. Gat, Optica 2016,
3, 189.
[39] B. G. Bale, K. Kieu, J. Kutz, F. Wise, Opt. Express 2009, 17, 23137.
[40] M. Horowitz, C. R. Menyuk, T. F. Carruthers, I. N. Duling, J. Lightwave
Technol. 2000, 18, 1565.
[41] J. M. Soto-Crespo, M. Grapinet, P. Grelu, N. Akhmediev, Phys. Rev. E
2004, 70, 066612.
[42] J. N. Kutz, B. Sandstede, Opt. Express 2008, 16, 636; T. F. Carruthers,
I. N. Duling, M. Horowitz, C. R. Menyuk, Opt. Lett. 2000, 25, 153.
[43] M. Y. Jeon, H. K. Lee, J. T. Ahn, K. H. Kim, D. S. Lim, E. H. Lee, Opt.
Lett. 1998, 23, 855.
[44] E. M. Dianov, A. V. Luchnikov, A. N. Pilipetskii, A. M. Prokhorov, Appl.
Phys. B: Photophys. Laser Chem. 1992, 54, 175.
Laser Photonics Rev. 2019, 13, 1800333 1800333 (9 of 9)
www.lpr-journal.org
[45] G. Sobon, K. Krzempek, P. Kaczmarek, K. M. Abramski, M. Nikodem,
Opt. Commun. 2011, 284, 4203.
[46] A. Grudinin, D. Richardson, D. Payne, Electron. Lett. 1993, 29, 1860.
[47] O. Gat, A. Gordon, B. Fischer, Phys. Rev. E 2004, 70, 046108; O. Gat,
A. Gordon, B. Fischer, New J. Phys. 2005, 7, 151.
[48] A. Gordon, B. Fischer, Phys. Rev. Lett. 2002, 89, 103901; A. Gordon,
B. Fischer, Opt. Lett. 2003, 28, 1326.
[49] E. Baumann, F. R. Giorgetta, J. W. Nicholson, W. C. Swann, I. Cod-
dington, N. R. Newbury, Opt. Lett. 2009, 34, 638.
[50] Y. C. Tong, L. Y. Chan, H. K. Tsang, Electron. Lett. 1997, 33, 983; M. A.
Muriel, J. Azaña, A. Carballar, Opt. Lett. 1999, 24, 1.
[51] G. Herink, F. Kurtz, B. Jalali, D. R. Solli, C. Ropers, Science 2017, 356,
50.
[52] D. R. Solli, C. Ropers, P. Koonath, B. Jalali, Nature 2007, 450, 1054.
[53] X. Wei, B. Li, Y. Yu, C. Zhang, K. Tsia, Opt. Express 2017, 25, 29098.
[54] P. Ryczkowski, M. Närhi, C. Billet, J. M. Merolla, G. Genty, J. M. Dud-
ley, Nat. Photonics 2018, 12, 221.
[55] A. F. J. Runge, N. G. R. Broderick, M. Erkintalo, Optica 2015, 2, 36.
[56] J. K. Jang, M. Erkintalo, S. G. Murdoch, S. Coen, Nat. Photonics 2013,
7, 657.
[57] O. Svelto, Principles of Lasers, Springer, New York 2010.
[58] G. P. Agrawal, Nonlinear Fiber Optics, Academic Press, New York
2007.
[59] G. Herink, B. Jalali, C. Ropers, D. R. Solli, Nat. Photonics 2016, 10,
321.
[60] X. Liu, Y. Cui, Advanced Photonics 2019, 1, 016003.
[61] J. M. Soto-Crespo, N. Akhmediev, Phys. Rev. Lett. 2005, 95, 024101.
[62] M. S. Kang, A. Nazarkin, A. Brenn, P. St. J. Russell, Nat. Phys. 2009,
5, 276.
[63] W. Ng, V. Stasyuk, Electron. Lett. 2018, 54, 303.
[64] K. Bagnell, A. Klee, P. J. Delfyett, J. J. Plant, P. W. Juodawlkis, Opt. Lett.
2018, 43, 2396.

C 2019 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
 PHYSICAL REVIEW LETTERS 123, 093901 (2019)

Revealing the Transition Dynamics from Q Switching to Mode Locking in a Soliton Laser
Xueming Liu,1,4,5,6,* Daniel Popa,2 and Nail Akhmediev3
1
State Key Laboratory of Modern Optical Instrumentation, College of Optical Science and Engineering, Zhejiang University,
Hangzhou 310027, China
2
Department of Engineering, University of Cambridge, Cambridge CB3 0FA, United Kingdom
3
Optical Sciences Group, Research School of Physics and Engineering, The Australian National University,
Canberra 2600, Australian Capital Territory, Australia
4
Institute for Advanced Interdisciplinary Research, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China
5
College of Astronautics, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China
6
School of Physics and Electronic Science, Hunan University of Science and Technology, Xiangtan 411201, China
(Received 5 February 2019; revised manuscript received 18 May 2019; published 27 August 2019)

Q switching (QS) and mode locking (ML) are the two main techniques enabling generation of ultrashort
pulses. Here, we report the first observation of pulse evolution and dynamics in the QS-ML transition stage,
where the ML soliton formation evolves from the QS pulses instead of relaxation oscillations (or quasi-
continuous-wave oscillations) reported in previous studies. We discover a new way of soliton buildup in an
ultrafast laser, passing through four stages: initial spontaneous noise, QS, beating dynamics, and ML. We
reveal that multiple subnanosecond pulses coexist within the laser cavity during the QS, with one dominant
pulse transforming into a soliton when reaching the ML stage. We propose a theoretical model to simulate
the spectrotemporal beating dynamics (a critical process of QS-ML transition) and the Kelly sidebands of
the as-formed solitons. Numerical results show that beating dynamics is induced by the interference
between a dominant pulse and multiple subordinate pulses with varying temporal delays, in agreement with
experimental observations. Our results allow a better understanding of soliton formation in ultrafast lasers,
which have widespread applications in science and technology.

DOI: 10.1103/PhysRevLett.123.093901

Ultrafast fiber lasers are increasingly used in a variety of
applications, ranging from optical communications to
sensors, medicine, and industry [1–9]. Their success is
partly due to their ability to generate pulses with a wide
range of parameters. Q switching (QS) and mode locking
(ML) are the two main techniques used for pulse generation
[10–14]. QS is a technique of modulation of the quality
factor (Q) of a laser cavity, where a pulse is formed when Q
is high, thus allowing the stored energy to be released
(lasing). This mechanism enables the formation of pulses
with durations ranging from micro- to nanoseconds and
repetition rates, typically around kilohertz, related to the
lifetime of the gain medium [11]. On the other hand, ML
enables the formation of a pulse by inducing a fixed phase
relation (synchronization) among the oscillating modes of a
laser cavity. Such mode interference leads to the formation
of pulses with durations ranging from tens of picoseconds
to sub-10 fs and repetition rates, typically around mega-
hertz, given by the inverse of the cavity round-trip time
(T R ) [12]. Because of their shorter durations and higher
peak intensities, in comparison to the QS pulses, ML pulses
dictate stricter requirements to parameters of laser cavity. A
common technique is a passive ML with an intracavity
pulse (soliton) being shaped through a balance of cavity
dispersion and pulse-triggered nonlinear effects [4,12].
Soliton lasers are being extensively investigated due to
their highly stable operation and excellent pulse quality,
placing them at the heart of ultrafast laser technology
[1,15,16]. However, despite an increasing research effort
dedicated to implementing QS and ML techniques in a
variety of laser configurations [9,17–19], most pulse
evolution and dynamics studies conducted in these systems
are theoretical rather than experimental. Consequently,
little is known about their real-time ultrafast transition
dynamics [20–24].
The time-stretch dispersive Fourier transform (TSDFT)
method is an emerging technique for studies of real-time
transient processes in nonlinear optical systems. Here the
spectrum of a broadband optical pulse is mapped, through
dispersion, into a temporal waveform that can be sub-
sequently digitized and processed in real time [25–31]. The
TSDFT technique has been successfully employed to study
the physical mechanisms of many important nonlinear
phenomena in ultrafast optics. These include the transient
dynamics of soliton molecules [28,32,33], internal motion
of dissipative solitons [34,35], optical rogue waves [36],
soliton explosions [37], and modulation instability in fiber
lasers [38]. More recent experimental observations based
on TSDFT techniques have shown that ML solitons appear
from dominant picosecond fluctuations in a narrowband

0031-9007=19=123(9)=093901(6) 093901-1 © 2019 American Physical Society

 PHYSICAL REVIEW LETTERS 123, 093901 (2019)
FIG. 1. Buildup process of solitons via QS. (a) Schematic
diagram showing the transition from initial noise to QS, follow-
up beating dynamics, and ML. T S , start-up time; T R , round-trip
time of laser cavity; T Q , time delay between two adjacent QS
pulses (i.e., lasing spikes). (Inset) Time enlargement of ML stage.
The pumping strength (red curve) increases rapidly (∼0.1 ms)
and remains constant afterwards. (b) Conceptual representation
of the four soliton buildup stages. The QS pulse is 3–4 orders of
magnitude longer than the ML soliton. Multiple pulses coexist
during the beating dynamics stage, with one dominant pulse
evolving into a soliton in the stable ML stage. (c) Experimental
results of the entire buildup process of laser solitons. The time
interval marked with red M is shown in detail in Fig. 3. The QS
stage is characterized by many (e.g., ∼5 × 105 here) lasing spikes
compared to a few (e.g., six in Ref. [32]) in the relaxation
oscillation stage. Arbitrary units is denoted by a.u.
collection of many similar fluctuations [26,33], suggesting
that they are formed from relaxation oscillations [32] and
spectral beating dynamics [26]. However, beating dynam-
ics is observed in the spectral domain only [26,32], and a
theoretical model describing the physical mechanisms
governing such dynamics is still missing.
By subtly designing the laser cavity and its components,
here we report the first direct observation of soliton
formation dynamics and its evolution in the transition
phase from QS to ML. Using a TSDFT technique, we
observe a new soliton buildup process, where solitons are
generated via QS rather than from relaxation oscillations
(or quasi-continuous-wave oscillations) as previously
reported [26,32]. A schematic diagram of the transition,
its conceptual representation, and experimental evidence
are depicted in Figs. 1(a)–1(c), respectively. Four transition
093901-2
FIG. 2. Experimental setup containing a fiber laser oscillator for
generating ML solitons via QS and a TSDFT setting for real-time
measurements.
stages are shown, starting from initial spontaneous noise,
successively experiencing QS, beating dynamics, and ML.
Moreover, we propose a theoretical model to simulate the
soliton formation through beating dynamics in both spec-
tral and temporal domains.
The experimental setup, containing a fiber laser oscillator
and a TSDFT setting, is shown in Fig. 2. We use a ring-
configured oscillator composed of a polarization-insensitive
carbon nanotube saturable absorber (CNTSA), a 3.5-m-long
birefringent erbium-doped fiber (EDF), ∼8.5 m of single-
mode fiber (SMF), and a hybrid combiner (WTI) (composed
of a wavelength division multiplexer, a tap coupler, and an
isolator). The EDF, with 6 dB=m absorption at 980 nm, is
used as a gain medium. The WTI is used to ensure the
unidirectional operation as the laser output and as the
input for the pump [provided by a laser diode (LD)].
The oscillator has a fundamental frequency (repetition rate)
of ∼16.956 MHz, corresponding to a T R ∼ 58.975 ns. The
real-time and time-averaged spectral data are recorded with a
high-speed real-time oscilloscope via a photodetector (PD)
and an optical spectrum analyzer (OSA), respectively. The
TSDFT technique is implemented by temporally stretching
the pulses through a 5.1-km-long dispersion-compensating
fiber (DCF) with a dispersion of ∼ − 160 ps=ðnm kmÞ.
The QS-ML transition phase is quite sensitive to the
characteristics of CNTSA and the laser cavity length Lcavity .
Here, Lcavity ≈ 12 m and the CNTSA film has a modulation
depth of ∼10% and a nonsaturable loss of ∼60%. When
Lcavity > ∼25 m, the QS-ML transition vanishes, while the
multisoliton operation appears. When the modulation depth
and the nonsaturable loss of CNTSA are ∼12% and ∼50%,
respectively, the relaxation oscillation occurs instead of QS.
To achieve the QS-ML transition phase, therefore, the laser
cavity together with CNTSA has to be designed accurately.
Figure 1(c) shows the experimental results recorded over
the entire buildup process of solitons from spontaneous
noise to ML via QS. The pump power is connected to the
laser cavity at time T ¼ 0. The population inversion
increases in the gain medium when 0 < T < ∼6.7 ms
(T S in Fig. 1). The photon number defined by the quantum
field fluctuations initially remains low [10]. Stimulated
emission becomes dominant afterwards, resulting in
 PHYSICAL REVIEW LETTERS 123, 093901 (2019)

FIG. 3. Experimental real-time data of the soliton buildup

process in the QS-ML transition phase, corresponding to the
time interval M in Fig. 1(c). The shot-by-shot experimental data
before and after the transition phase illustrate the temporal and
spectral information, respectively. The intensity profile of pulses
(z axis) evolves along with the round-trips (y axis) and the
intracavity time (x axis). The blue projection on the y and z plane
shows the intensity profile along with the round-trips, with a
close-up (marked as range A) shown in Fig. 4 for details.

generation of QS pulses. Note that the spacing between two

adjacent pulses, T Q ≈ 0.12 ms, is uniform in the QS stage.
In contrast, the spacing is nonuniform in Refs. [10,32],
which is clear evidence of relaxation oscillations. A QS
stage usually lasts for tens of seconds [∼1 min . in
Fig. 1(c)] with the pulsing period being random.
Conversely, the relaxation oscillations with lasing spikes
last less than 1 ms [32]. The comparison and differences
between the QS and relaxation oscillations are presented in
Fig. S1 in the Supplemental Material [39]. We have found
that both T S and T Q are dependent on the pump power,
shown in Fig. S2 in the Supplemental Material [39].
We analyze the shot-by-shot experimental data by
segmenting the TSDFT recorded time series with a T R ¼
58.975 ns periodicity. The time interval M in Fig. 1(c) is FIG. 4. Details of the transition phase from QS to ML.
redrawn as a spectrotemporal picture, shown in Fig. 3. The (a) Magnification of the region A in Fig. 3. (b) Magnification
x axis in this plot depicts the time within a single round-trip of the region B in (a), showing a detailed beating pattern and
(i.e., from 0 to 58.975 ns), while the y axis shows the the growth of Kelly sidebands. (c) The cross section of x axis in
dynamics across consecutive round-trips. Multiple subna- (a) at 34.44 ns intracavity time. (Inset) Beating dynamics
nosecond pulses with different peak powers coexist in the magnification. (d) The cross section of y axis in (a) at the
cavity during this stage, as shown in Fig. 4(d) and Fig. S3 round-trip K. It shows the competing of the dominant pulse with
(see Supplemental Material [39]). multiple subordinate pulses denoted as P1–P6. (e) Optical
The A region in Fig. 3 (i.e., QS-ML transition phase) is spectrum of solitons recorded by OSA (red curve) and an
exemplary single-shot spectrum (black curve) recorded by our
magnified in Fig. 4(a) and further magnified in Fig. 4(b)
high-speed real-time oscilloscope (TSDFT setting), correspond-
(B region), in order to see clearly the beating pattern. Kelly ing to the last frame in (a).
sidebands do not appear during the first 40 round-trips [see
Fig. 4(b)] of the QS-ML transition phase. They pop up
together with the beating pattern. Multiple sidebands grow stage as Fig. 4(b) shows. This process is depicted in
gradually and simultaneously. These features are character- Figs. 4(c) and 4(d) for cross sections marked at 34.44 ns
istic of solitons [6]. They are not formed at the early stage along the x axis and K along the y axis in Fig. 4(a). A
of the transition phase, although the QS pulses exist at this modulation pattern, which is a result of beating dynamics,

093901-3
 PHYSICAL REVIEW LETTERS 123, 093901 (2019)

can be seen in the inset of Fig. 4(c). Multiple subnano-

second pulses appear at the QS stage, as shown in Fig. 4(d)
and in Fig. S3 of the Supplemental Material [39], with one
dominant pulse eventually evolving into a stationary soliton
in the ML stage [Fig. 4(a)]. Using the TSDFT technique,
femtosecond pulses can be broadened to nanosecond
pulses, while pulses over tens of picoseconds are hardly
broadened [32]. As a result, subnanosecond pulses do exist
in the first 40 round-trips in Fig. 4(b), while femtosecond
ML solitons appear only in the subsequent round-trips. A
full animation of this evolution in the real-time exper-
imental observation of the QS-ML transition phase is
shown in the video of the Supplemental Material [44].
Figure 4(e) shows the soliton optical spectrum (red
curve) and an exemplary single-shot spectrum (black
curve), recorded by the OSA and the high-speed real-time
oscilloscope, respectively. The two curves (including Kelly
sidebands) are consistent, giving a Δλ ¼ 35 nm bandwidth
for the optical spectrum mapped (stretched) over a Δt ¼
28.6 ns time spacing. The mapping relation between
spectral and temporal domains can be expressed as Δt ¼
jDjLΔλ [29,32], where D is the dispersion parameter and L
is the length of the DCF, with D ∼ −160 ps=ðnm kmÞ and
L ∼ 5100 m (see Fig. 1) in our experiments. If the TSDFT
technique is not used, the beating phenomenon and Kelly
sidebands cannot be discovered, just as in the previous
report [32].
To better understand the soliton dynamics and its
formation, we perform numerical simulations based on a
round-trip circulating-pulse method [40]. To discover the
beating behavior, we consider the influence of the nonlinear
refractive index defined by the second term of Eq. (1),
which can cause the temporal delays of pulses. We use a FIG. 5. Numerical simulations revealing the soliton formation
modified nonlinear Schrödinger equation (MNSE) for through beating dynamics. (a) Spectral and (b) temporal evolu-
simulating a propagating pulse through an optical fiber, tions in the soliton buildup process. The phenomena of spectral
beating and Kelly sidebands are clearly seen. Multiple (two in this
given by (see the Supplemental Material [39] for the
case) subordinate pulses play key roles in the beating pattern. The
detailed derivation) disappearance of subordinate pulses will terminate the beating
dynamics described as auxiliary-pulse ML in Ref. [26]. (c) Spec-
∂A γλ0 jAj2 ∂A iβ2 ∂ 2 A β3 ∂ 3 A tral and (d) temporal profiles of stationary solitons.
þ þ −
∂z 2πc ∂T 2 ∂T 2 6 ∂T 3
g g ∂2A is the pulse energy, and Es is the gain saturation energy
¼ A þ iγjAj2 A þ 2 2 : ð1Þ
2 2Ωg ∂T [40,41]. In our simulations, we start with an initial small
signal as noise background, as shown in Fig. S5 in the
Here A, λ0 , c, and γ represent the pulse electric field Supplemental Material [39]. This initial signal is nearly 8
envelope, its central wavelength, the speed of light in orders of magnitude weaker than the stable soliton. To
vacuum, and the fiber cubic refractive nonlinearity, respec- match the experimental conditions, we use the following
tively. β2 and β3 are the fiber second- and third-order parameters: Ωg ¼ 40 nm, β3 ¼ 0.3 ps3 =km, λ0 ¼ 1565 nm,
dispersion coefficients, respectively. To simplify the Es ¼ 90 pJ; β2 ¼ 13.5 ps2 =km and γ ¼ 1.8 W−1 km−1 for
MNSE, a frame of reference moving with the pulse at EDF; β2 ¼ −21.7 ps2 =km and γ ¼ 1 W−1 km−1 for SMF.
the group velocity vg is used by making the transformation The spectral and temporal evolutions of soliton forma-
T ¼ t − z=vg [41], where t and z are the time and the tion are shown in Figs. 5(a) and 5(b). The stable solutions at
propagation distance variables, respectively. We use g ¼ 700 round-trips are shown in Figs. 5(c) and 5(d) in the
g0 expð−Ep =Es Þ to describe the EDF gain (note that g ¼ 0 spectral and temporal domains, respectively. Our numerical
for SMF), where g0 is the small-signal gain coefficient, Ep model clearly reveals the spectral beating behavior and

093901-4
 PHYSICAL REVIEW LETTERS 123, 093901 (2019)
Kelly sidebands [see Fig. 5(a)]. Like in the experimental
observations, spectra are narrow and no Kelly sideband
appears at the early stage of beating dynamics (below 100
round-trips). They broaden rapidly afterwards with Kelly
sidebands increasing simultaneously. The sidebands are
asymmetrical due to the third-order dispersion term β3
[see Eq. (1)]. The numerical results are in good agreement
with the experimental observations shown in Fig. 4(b).
To simplify MNSE, the EDF birefringence is omitted
in Eq. (1).
Our numerical simulations confirm that the beating
pattern phenomenon originates from the interference of
multiple pulses with varying temporal delays induced by
the nonlinear refractive index and the dispersion. The
second term in Eq. (1) plays an important role in the
occurrence of the beating pattern, which can induce varying
pulse delay. The spectral amplitude can be obtained, via
Fourier transform, from the temporal amplitude of all the
pulses accompanied by their respective temporal delays. As
a result, the pattern in the spectral regime demonstrates the
beating behavior. Numerical results show that beating
dynamics occurs in the region from ∼40 to 170 round-
trips [see Figs. 5(a) and 5(b)], due to the temporal delays of
these pulses. A dominant pulse is seen together with
multiple (two in this case) subordinate pulses during the
beating dynamics process, with the disappearance of the
subordinate pulses terminating such process. The dominant
pulse has a relatively low power and large duration during
the early stage (∼40–80 round-trips) of the buildup
process [see Fig. 5(b)], corresponding to a sharp Fourier
transformed spectral peak with narrow bandwidth [see
Fig. 5(a)]. Such pulse then increases its power and shortens
its duration along with the round-trips and simultaneously
reduces its spectral peak and broadens its spectral band-
width. After ∼170 round-trips, the two subordinate pulses
die away [see Fig. 5(b)] and the spectral beating pattern
vanishes [see Fig. 5(a)].
The numerical calculations show that both the nonlinear
refractive index and the dispersion can cause the time delay
of respective pulses. When the nonlinear refractive index is
excluded from Eq. (1), the beating pattern [see Fig. 5(a)]
will be weakened and blurred. In this case, the numerical
results deviate from the experimental observations.
In conclusion, by using a TSDFT technique, we reveal
the real-time soliton buildup process in the ultrafast laser
cavity that transforms the initial spontaneous noise into
stable ML (via QS). We experimentally observe, for the
first time to our best knowledge, the single-shot dynamics
and evolution of solitons from QS to ML transition phase.
A new way of soliton buildup in lasers is discovered, where
solitons are generated by ML via QS, rather than from
relaxation oscillations or quasi-continuous-wave oscilla-
tions as previously reported [26,32]. It is revealed that
multiple subnanosecond pulses coexist in the laser cavity
during the QS stage, with one dominant pulse transitioning
093901-5
into a soliton of the stable ML stage. We propose a
theoretical model to reveal the soliton formation through
the QS-ML transition phase (characterized as beating
dynamics), thus validating the experimental results. The
numerical results show that a dominant pulse together with
multiple subordinate pulses induces the spectral beating
dynamics, which plays the key role in the buildup process of
solitons. Both theoretical and experimental results confirm
that the beating behavior is an inevitable phenomenon in the
formation of mode-locked lasers [26,32,39]. Our findings
open new perspectives for ultrafast transient dynamics and
pathways of pulse evolution and will bring new insights into
the design and application of lasers.
We thank Y. Zhang, X. Yao, C. Jin, and J. Zhen for
fruitful discussions. This Letter was partially supported by
the National Natural Science Foundation of China under
Grants No. 61525505 and No. 11774310.
*
liuxueming72@yahoo.com
[1] U. Keller, Nature (London) 424, 831 (2003).
[2] L. G. Wright, D. N. Christodoulides, and F. W. Wise,
Science 358, 94 (2017).
[3] M. Pang, W. He, X. Jiang, and P. S. J. Russell, Nat.
Photonics 10, 454 (2016).
[4] P. Grelu and N. Akhmediev, Nat. Photonics 6, 84 (2012).
[5] J. M. Dudley, F. Dias, M. Erkintalo, and G. Genty, Nat.
Photonics 8, 755 (2014).
[6] H. A. Haus and W. S. Wong, Rev. Mod. Phys. 68, 423
(1996).
[7] D. V. Churkin, S. Sugavanam, N. Tarasov, S. Khorev, S. V.
Smirnov, S. M. Kobtsev, and S. K. Turitsyn, Nat. Commun.
6, 7004 (2015).
[8] A. Klein, G. Masri, H. Duadi, K. Sulimany, O. Lib, H.
Steinberg, S. A. Kolpakov, and M. Fridman, Optica 5, 774
(2018).
[9] B. Oktem, C. Ülgüdür, and F. Ö. Ilday, Nat. Photonics 4,
307 (2010).
[10] O. Svelto, Principles of Lasers (Plenum, New York, 1998).
[11] D. Popa, Z. Sun, T. Hasan, F. Torrisi, F. Wang, and A. C.
Ferrari, Appl. Phys. Lett. 98, 073106 (2011).
[12] G. P. Agrawal, Applications of Nonlinear Fiber Optics
(Academic Press, San Diego, 2008).
[13] S. A. Kolpakov, H. Kbashi, and S. V. Sergeyev, Optica 3,
870 (2016).
[14] C. M. Eigenwillig, W. Wieser, S. Todor, B. R. Biedermann,
T. Klein, C. Jirauschek, and R. Huber, Nat. Commun. 4,
1848 (2013).
[15] M. E. Fermann and I. Hartl, Nat. Photonics 7, 868 (2013).
[16] J. Gabzdyl, Nat. Photonics 2, 21 (2008).
[17] K. Sulimany, O. Lib, G. Masri, A. Klein, M. Fridman, P.
Grelu, O. Gat, and H. Steinberg, Phys. Rev. Lett. 121,
133902 (2018).
[18] D. Chaparro and S. Balle, Phys. Rev. Lett. 120, 064101
(2018).
[19] A. H. Quarterman, K. G. Wilcox, V. Apostolopoulos, Z.
Mihoubi, S. P. Elsmere, I. Farrer, D. A. Ritchie, and A.
Tropper, Nat. Photonics 3, 729 (2009).
 PHYSICAL REVIEW LETTERS 123, 093901 (2019)
[20] M. Kues, C. Reimer, B. Wetzel, P. Roztocki, B. E. Little,
S. T. Chu, T. Hansson, E. A. Viktorov, D. J. Moss, and R.
Morandotti, Nat. Photonics 11, 159 (2017).
[21] W. Liang, V. S. Ilchenko, A. A. Savchenkov, A. B. Matsko,
D. Seidel, and L. Maleki, Phys. Rev. Lett. 105, 143903
(2010).
[22] M. Leonetti, C. Conti, and C. Lopez, Nat. Photonics 5, 615
(2011).
[23] X. Liu, Phys. Rev. A 81, 023811 (2010).
[24] A. Chong, L. Wright, and F. Wise, Rep. Prog. Phys. 78,
113901 (2015).
[25] K. Goda, K. K. Tsia, and B. Jalali, Nature (London) 458,
1145 (2009).
[26] G. Herink, B. Jalali, C. Ropers, and D. R. Solli, Nat.
Photonics 10, 321 (2016).
[27] A. Tikan, S. Bielawski, C. Szwaj, S. Randoux, and P. Suret,
Nat. Photonics 12, 228 (2018).
[28] G. Herink, F. Kurtz, B. Jalali, D. R. Solli, and C. Ropers,
Science 356, 50 (2017).
[29] Y. C. Tong, L. Y. Chan, and H. K. Tsang, Electron. Lett. 33,
983 (1997).
[30] M. A. Muriel, J. Azaña, and A. Carballar, Opt. Lett. 24, 1
(1999).
[31] C. Lei and K. Goda, Nat. Photonics 12, 190 (2018).
[32] X. Liu, X. Yao, and Y. Cui, Phys. Rev. Lett. 121, 023905
(2018).
[33] J. Peng and H. Zeng, Laser Photonics Rev. 12, 1800009
(2018).
[34] K. Krupa, K. Nithyanandan, U. Andral, P. Tchofo-Dinda,
and P. Grelu, Phys. Rev. Lett. 118, 243901 (2017).
[35] P. Ryczkowski, M. Närhi, C. Billet, J. M. Merolla, G. Genty,
and J. M. Dudley, Nat. Photonics 12, 221 (2018).
093901-6
[36] D. R. Solli, C. Ropers, P. Koonath, and B. Jalali, Nature
(London) 450, 1054 (2007).
[37] A. F. J. Runge, N. G. R. Broderick, and M. Erkintalo, Optica
2, 36 (2015).
[38] D. R. Solli, G. Herink, B. Jalali, and C. Ropers, Nat.
Photonics 6, 463 (2012).
[39] See Supplemental Material at http://link.aps.org/
supplemental/10.1103/PhysRevLett.123.093901 for more
detailed discussions, which includes Refs. [10,11,26,32,40–
43]. Section 1 compares the Q switching and relaxation
oscillations. Section 2 describes the start-up time and
repetition rate of Q switching versus pump power. Section 3
magnifies three lasing spikes preceding the formation of
stable ML solitons. Section 4 compares optical spectra of
QS and ML pulses. Section 5 describes the proposed
theoretical model in detail. Section 6 illustrates the initial
signal with the noise background. Section 7 demonstrates
the evolution from many pulses to a soliton via beating
behavior.
[40] X. Liu, Y. Cui, D. Han, X. Yao, and Z. Sun, Sci. Rep. 5,
9101 (2015).
[41] G. P. Agrawal, IEEE Photonics Technol. Lett. 2, 875
(1990).
[42] B. Dong, C. Liaw, J. Hao, and J. Hu, Appl. Opt. 49, 5989
(2010).
[43] G. P. Agrawal, Nonlinear Fiber Optics (Academic Press,
San Diego, 2007).
[44] See Supplemental Material at http://link.aps.org/
supplemental/10.1103/PhysRevLett.123.093901 for a video
showing the experimental real-time measurements from
DCF (i.e., with time-stretch technique) for the transition
dynamics.
 PHYSICAL REVIEW LETTERS 121, 023905 (2018)

Real-Time Observation of the Buildup of Soliton Molecules

Xueming Liu,1,2,* Xiankun Yao,1 and Yudong Cui1
1
State Key Laboratory of Modern Optical Instrumentation, College of Optical Science and Engineering,
Zhejiang University, Hangzhou 310027, China
2
State Key Laboratory of Transient Optics and Photonics, Xi’an Institute of Optics and Precision Mechanics,
Chinese Academy of Sciences, Xi’an 710119, China

(Received 13 February 2018; published 12 July 2018)

Real-time spectroscopy access to ultrafast fiber lasers opens new opportunities for exploring complex
soliton interaction dynamics. Here, we have reported the first observation, to the best of our knowledge,
of the entire buildup process of soliton molecules (SMs) in a mode-locked laser. We have observed that
the birth dynamics of a stable SM experiences five different stages, i.e., the raised relaxation oscillation
(RO) stage, beating dynamics stage, transient single pulse stage, transient bound state, and finally the stable
bound state. We have discovered that the evolution of pulses in the raised RO stage follows a law that only
the strongest one can ultimately survive and, meanwhile, the pulses periodically appear at the same
temporal positions for all lasing spikes during the same RO stage (named as memory ability) but they lose
such ability between different RO stages. Moreover, we have found that the buildup dynamics of SMs is
quite sensitive to both the polarization state of intracavity light and the fluctuation of pump power. These
results provide new perspectives into the ultrafast transient process in mode-locked lasers and the dynamics
of complex nonlinear systems.

DOI: 10.1103/PhysRevLett.121.023905

Dispersion is responsible for pulse broadening or com-
pression that is dependent on the dispersion value [1–3].
Resulting from an elegant balance between dispersion and
nonlinearity [3–6], however, the optical soliton has remark-
able robustness over long-distance propagation. With a
particlelike nature, the solitons exhibit profound nonlinear
optical dynamics and excitation ubiquitous in many fields
of physics [7–9], including fluids [10], plasmas [11], fibers
[12], optical systems [13], complex networks [14], and
Bose-Einstein condensates [15]. The mode-locked fiber
laser constitutes an ideal test bed for investigating ultrashort
pulse dynamics [16], where such pulses arise from the
dynamical balance among nonlinearity, dispersion, and
environmental energy exchange and then are referred to
as dissipative solitons [7,17]. As several solitons coexist
in a laser cavity, they can constitute the bound state, which
is frequently referred to as the soliton molecule (SM)
[1,8,18–21].
From the 1990s, the starting dynamics of mode-locked
lasers had been reported experimentally and theoretically
[22–24], yet the spectral dynamics during the buildup of
pulse lasers has not been measured directly [1]. Recently,
the dynamic behavior of mode-locked lasers has been
investigated via the time-stretch dispersive Fourier trans-
form (TS DFT) technique [25], in which the spectral
information is mapped into the time domain [1,26,27].
This new technique opens new opportunities for exploring
SM dynamics and then real-time observations of internal
motion and the complex interaction dynamics of SMs in
mode-locked lasers have been demonstrated [1,3].
Furthermore, transient coherent multisoliton states have
been observed experimentally [17], where the short-lived
SMs grow from noise and rapidly decay. However, the
entire buildup process of stable long-lived SMs has not
been discovered so far. Here, we demonstrate the first direct
observation of the entire buildup process of stable SMs in a
mode-locked fiber laser by means of the TS DFT technique.
Our measurements reveal very complicated dynamics in
the birth of SMs, e.g., the raised RO stage, beating
behavior, Kelly sideband, spectrum induced by four-wave
mixing (FWM), transient single soliton, and transient
bound state.
Figure 1 illustrates the experimental setup. The laser
system consists of a carbon nanotube saturable absorber
(CNT SA), a segment of erbium-doped fiber (EDF), a
polarization controller (PC), some single-mode fiber (SMF)
pigtails, and a polarization-independent hybrid combiner of
a wavelength division multiplexer, a tap coupler, and an
isolator (WTI). The gain medium is a 3.5-m-long EDF with
6 dB=m peak absorption at 980 nm, pumped by a laser
diode (LD). The CNT film acts as a modelocker to initiate
the soliton operation. The PC and optical chopper are used
to optimize the mode-locking performance and control the
pump power, respectively. The dispersions of SMF and
EDF are −22 and 11.6 ps2 =km at 1.55 μm, respectively,
resulting in a net cavity dispersion of −0.06 ps2 . The total
cavity length ∼7.8 m entails a fundamental repetition rate
of 26.3 MHz. The time-averaged spectral and real-time

0031-9007=18=121(2)=023905(6) 023905-1 © 2018 American Physical Society

 PHYSICAL REVIEW LETTERS 121, 023905 (2018)

FIG. 1. Schematic diagram of the experimental setup for the mode-locked laser.

temporal detections are recorded with an optical spectrum Figure 2(a) shows that, after the zero round-trip number,
analyzer (OSA) and a high-speed real-time oscilloscope the oscilloscope records the spectral characterization of
(20 Gsamples=s sampling rate) together with a 25-GHz- SMs. For example, Kelly sidebands appear at the wave-
bandwidth photodetector (PD), respectively. The DFT is lengths of about 1547 and 1578 nm, the spectra induced by
implemented by temporally stretching the solitons in a
5-km-long dispersion-compensating fiber (DCF) with
dispersion of about −160 ps=ðnm kmÞ.
The mode-locked laser here emits the continuous wave,
single soliton, and SMs at the pump power of ∼10, ∼16,
and ∼20 mW, respectively. Figures 2(a) and 2(b) show the
recorded results over the entire buildup process of SMs in a
mode-locked laser with and without the TS DFT technique,
respectively. The recorded time series are segmented with
respect to the round-trip time and then the buildup
dynamics of solitons can be depicted by the round-trip
time and the round-trip number. The y and x axes depict the
time within a single round-trip (i.e., from 0 to ∼38 ns) and
the dynamics across consecutive round-trips, respectively.
For the convenience of reference, we set the beginning time
of the single-pulse mode-locking state as the zero round-
trip number. The red curves in the insets of Figs. 2(a) and
2(b) denote the cross sections at the round-trip numbers of
−570 and −590, respectively. The blue curves in the insets
of Figs. 2(a) and 2(b) represent the cross sections at the
intracavity time of 34.5 ns. An enlargement of RO stage
and the corresponding cross sections in Fig. 2(a) are shown
in Figs. S1(a)–S1(c) in the Supplemental Material [28].
Figure 2 demonstrates that a raised RO stage with six
spikes occurs prior to the zero round-trip number and then
the beating behavior, Kelly sidebands, and transient bound
state appear before the stable SMs. The direct measurement
only exhibits the information of solitons in the temporal
domain, as shown in Fig. 2(b). A movie illustrates the FIG. 2. Experimental real-time characterization of the entire
experimental real-time observation in detail (see the buildup process of a SM in a mode-locked laser (a) with and (b)
Supplemental Material movie at Ref. [29]). Note that, in without the TS DFT technique, respectively. The intensity profile
evolves along with the round-trip time (y axis) and the round-trip
Figs. 2(a) and 2(b), the difference of the RO stage with and number (x axis). Insets in (a),(b) denote the cross sections at the
without the TS DFT technique originates from the reso- −570 and −590 round-trip numbers (red curve) and at the time
lution of the oscilloscope. After using the TS DFT (blue curve) of 34.5 ns, respectively. Indications for P0 to P5
technique, the picosecond or subpicosecond pulses are represent a pulse appearing on six spikes at the time of 34.5 ns.
broadened to subnanosecond pulses [see Fig. S2(b) in the An enlargement of the raised RO stage in (a) is shown in Fig. S1
Supplemental Material [28] for details]. in the Supplemental Material [28].

023905-2
 PHYSICAL REVIEW LETTERS 121, 023905 (2018)

FWM occur near 3500 round-trip, and the spectral modu-

lation arises in the bound-state stage. See the Supplemental
Material [30] for a full animation. All of these features
cannot be observed in the direct measurement (see the
Supplemental Material [29]). In contrast, prior to the zero
round-trip number, the oscilloscope records the temporal
information of SMs. A detailed description and an analysis
for resolving the spectral and temporal information are
given in the Supplemental Material [28]. The existence of a
RO stage is a typical characteristics of the transient
behavior of lasers [2,31,32]. Multiple pulses coexist in
the laser cavity during this stage, where a typical example is
shown in the inset of Fig. 2(a) (see the red curve) or
Fig. S1(c) in the Supplemental Material [28]. The exper-
imental observation demonstrates that multiple subnano-
second pulses appear in the RO stage but only one
dominant pulse [see the insets of Figs. 2(a) and 2(b)]
gradually evolves into the stationary mode-locking soliton.
As a result, only the strongest pulse can survive ultimately
and the others die.
The experimental results exhibit an interesting phenome-
non; i.e., during the RO stage, the pulses are able to
reappear at the same temporal positions for all lasing
spikes. For instance, a pulse is observed at a fixed intra-
cavity time of 34.5 ns for all six lasing spikes (see P0 to P5
in Fig. 2). Note that a fixed time in the y axis corresponds to
a certain position of the cavity for each round-trip.
The evolution of this pulse at the intracavity time of
34.5 ns over consecutive round-trips is shown in the insets
of Fig. 2 (see the blue curves) or Fig. S1(b) in the FIG. 3. Formation and evolution of a SM with beating
Supplemental Material [28]. Although this pulse disappears dynamics. (a) Experimental real-time interferograms during the
at the round-trips from about −9500 to −8400, −7500 to formation and evolution of SM. A full animation is shown in
−6500, −5500 to −4600, −3500 to −2700, and −1800 to movie of the Supplemental Material [30]. (b),(c) Close-ups of the
−1100, it reappears at the same relative position in the laser data from the A and B regions of (a), revealing the interference
cavity when it revives. Therefore, it seems that the pulses pattern for the beating dynamics.
are able to “remember” some properties before their
reappearance. This “memory ability” is a notable feature after another beating process, as shown in Fig. 3(c) [an
during the RO stage of mode-locked fiber lasers. expanded view of the B region in Fig. 3(a)]. Figures 3(b)
Using the TS DFT technique, we have recorded the shot- and 3(c) show that both of the two beating dynamics last
by-shot spectral information of the laser evolving [see over ∼800 round-trips. The full animation is demonstrated
Fig. 2(a)]. To reveal the formation and evolution of SMs, in the movie of the Supplemental Material [30]. During the
the planform of Fig. 2(a) is redrawn in Fig. 3. Figure 3(a) round-trips from ∼3200 to 3800, as shown in Fig. 3(a), the
exhibits the experimental observations in the real-time laser spectra are slightly broadened by the spontaneous and
series of interferograms, tracking the entire formation multiple FWM effect [33,34].
process of a stable SM. Figures 3(b) and 3(c) are enlarge- Via the second beating process, one single soliton is
ments of the A and B regions in Fig. 3(a), respectively. broken into two. In succession, the transient bound state
Figure 3(a) demonstrates that, after the RO stage, the starts where two newly generated solitons interact inten-
soliton evolution of the mode-locked laser experiences sively. The Fourier transform of each single-shot spectrum
different stages such as the transient single pulse, transient corresponds to the field autocorrelation of the momentary
bound state, and finally the stable SM. An obvious beating bound-state solitons [1], tracing the separation between the
behavior is observed between the RO stage and the two solitons, as shown in Fig. 4(a). The real-time data
transient single pulse stage, as shown in the A region of illustrate the complex temporal evolution; e.g., two solitons
Fig. 3(a) or its expanded view [see Fig. 3(b)]. Clear Kelly arise from a single soliton near the round-trip of 3500 and
sidebands appear in the transient single-pulse stage. The the stable bound state appears after the round-trip of ∼6000
transient single pulse evolves to the transient bound state (see also the Supplemental Material [30]). Figure 4(b)

023905-3
 PHYSICAL REVIEW LETTERS 121, 023905 (2018)

FIG. 4. (a) Field autocorrelation of each single-shot spectrum, exhibiting the separation between two solitons during the transient
bound-state region. (b) Dynamics of the SM formation mapped into the interaction plane from about 3600 to 6500 round-trips.

shows the dynamics of the SM formation mapped into the
interaction plane from about 3600 to 6500 round-trips. In
Fig. 4(b), the angle (α) represents the relative phase
between the two solitons and the radius (R) corresponds
to the separation between them. The achievement for α is
shown in the Supplemental Material [28]. Prior to the
establishment of the stable bound state, the trajectory
illustrates the evolution of the two solitons over periodic
metastable soliton separations. Finally, the system reaches
a fixed point with a locked relative phase, as shown in
Fig. 4(b), and then two solitons form a SM settled at a
constant binding separation together with a fixed phase
relationship.
Figures 5(a) and 5(b) illustrate the spectra of the laser
pulses measured by the OSA and the real-time TS DFT
technique, respectively. Two curves in Fig. 5(a) demon-
strate the same data with a linear or logarithmic scale.
Figure 5(b) exhibits the last frame in the real-time series
shown in Fig. 2(a). The experimental observations show
that the real-time records have evident Kelly sidebands,
which are typical characteristics of solitons in the presence
of periodic amplification [35–37]. The real-time single-shot
spectra measured by the TS DFT technique agree quite well
with the time-averaged optical spectra measured by the
OSA, highlighting the mapping relationship linked by the
dispersion [38]. Therefore, the TS DFT technique can
accurately map the spectral information of SMs [see
Fig. 5(a)] into the temporal domain [see Fig. 5(b)].
For simplicity, the specific mapping relation can be
expressed as [38,39]
Δt ¼ jDjLΔλ; ð1Þ
where Δλ is the bandwidth of the optical spectrum, Δt is the
time spacing into which the optical spectrum is mapped,
and D and L are the dispersion parameter and length of
DCF, respectively. Since D is about −160 ps=ðnm kmÞ and
L is 5 km in our experiments, the relation between the time
spacing and the wavelength can be expressed as
Δt ¼ 0.8 ns=nm × Δλ. An example is that Δt1 ¼ 12.6 ns
for Δλ1 ¼ 15.8 nm, where Δλ1 and Δt1 are the time
spacing and spectral bandwidth from the first Kelly side-
band to the central wavelength and central time, respec-
tively, as shown in Fig. 5. Obviously, the experimental
results show excellent agreement with the theoretical
estimations using the equation above.
The RO stage is the general characterization of the
transient behavior of lasers [2,31]. When the modelocker
(i.e., CNT SA) is excluded from the experiment setup, the
experimental results demonstrate that the laser starts at a
RO stage with a damped behavior (see Fig. 6), rather than a
raised behavior (see Fig. 2). Specifically, the laser without
a modelocker emits a uniform optical wave distributed
throughout the laser cavity (see Fig. 6), which is quite
different from the mode-locked laser where multiple pulses
are initiated at this stage [see Figs. 2(a) and 7(b)].

FIG. 5. (a) Optical spectra of SM measured by OSA. (b) Exemplary single-shot spectrum, corresponding to the last frame in Fig. 2(a).

023905-4
 PHYSICAL REVIEW LETTERS 121, 023905 (2018)
FIG. 6. RO stage of lasers without the modelocker. The
projection on the round-trip intensity plane shows the intensity
profile along with the round-trips.
In experiments, we have observed that the buildup
process of dissipative solitons is quite sensitive to external
perturbations such as the change of the light polarization
state and the fluctuation of the pump power. Figure 7(a)
demonstrates the buildup process of SMs in a laser mode-
locked by the nonlinear polarization rotation technique.
Because of the perturbation of the light polarization state,
before the stable SM formation this laser experiences a
complex transient process lasting over 2.3 × 105 round-
trips, which is more than 23 times longer than that of the
CNT-based mode-locked laser (see Fig. 2). Similar works
were reported in Refs. [17,40]. The laser mode-locked
by CNT SA is insensitive to the polarization change
FIG. 7. Buildup process of SMs in lasers mode-locked by (a)
the nonlinear polarization rotation technique and (b) CNT SA
with a fluctuating pump power. The projection shows the
temporal evolution along with the round-trips.
023905-5
[19,36,41], yet the fluctuation of pump power can influence
the birth of SMs. Figure 7(b) demonstrates a typical result
where there exist three RO stages rather than only the one
RO stage shown in Fig. 2. The pulses appear at the same
temporal positions for all spikes in each RO stage, yet they
occur on different positions between different RO stages. For
instance, the strongest pulse appears at the temporal posi-
tions of about 4.1, 19.7, and 23.5 ns for the first, second,
and third RO stages, respectively, as shown in Fig. 7(b).
Therefore, the pulses only have the memory ability at the
same RO stage but lose such ability between different RO
stages. In this Letter, the influence of polarization change in
the laser cavity has been overcome effectively by optimizing
the CNT SA design and the laser system and, meanwhile, the
fluctuation of pump power has also been decreased evi-
dently. The fluctuations of pump power for Figs. 2 and 7(b)
are shown in Figs. S5(a) and S5(b) in the Supplemental
Material [28], respectively. As a result, the entire buildup
process of SMs in mode-locked lasers is observed by using
the emerging TS DFT technique.
In summary, we have experimentally demonstrated the
evidence that the external perturbations (e.g., the change of
the light polarization state and the fluctuation of pump
power) can seriously influence the buildup process and
dynamics of SMs. By decreasing such perturbations and
optimizing the modelocker and the laser system, we have
successfully tracked the formation and evolution of SMs in
the cavity of mode-locked lasers using the TS DFT
technique. Therefore, we have experimentally observed
the real-time dynamics of the entire buildup process of
stable SMs for the first time to our best knowledge. These
observations show that the entire buildup process of SMs
usually experiences five different stages, i.e., the raised RO
stage, two times of the spectral beating dynamics stage,
transient single pulse stage, transient bound state, and
finally stable bound state. The mode-locked laser (i.e.,
with a modelocker) undergoes the raised RO stage where
multiple pulses occur; however, the laser without a mod-
elocker experiences the damped RO stage where no pulse
appears. We discover two interesting phenomena in mode-
locked lasers; i.e., the pulses in the raised RO stage evolve
with the rule of the survival of the strongest and they have
the memory ability at the same RO stage [see Fig. 2(a)] but
lose such ability between different RO stages [see
Fig. 7(b)]. We believe that our results can provide some
new perspectives into the ultrafast transient dynamics of
mode-locked lasers and will bring useful insights into laser
designs and applications.
We thank M. Pang, X. Han, G. Chen, W. Li, G. Wang,
and Y. Zhang for fruitful discussions. This Letter was
partially supported by the National Natural Science
Foundation of China under Grants No. 61525505,
No. 11774310, and No. 61705193 and by the Key
Scientific and Technological Innovation Team Project in
Shaanxi Province (2015KCT-06).
 PHYSICAL REVIEW LETTERS 121, 023905 (2018)
*
Corresponding author.
liuxueming72@yahoo.com
[1] G. Herink, F. Kurtz, B. Jalali, D. R. Solli, and C. Ropers,
Science 356, 50 (2017).
[2] O. Svelto, Principles of Lasers (Springer, New York, 2010).
[3] K. Krupa, K. Nithyanandan, U. Andral, P. Tchofo-Dinda,
and P. Grelu, Phys. Rev. Lett. 118, 243901 (2017).
[4] G. Herink, B. Jalali, C. Ropers, and D. R. Solli, Nat.
Photonics 10, 321 (2016).
[5] M. Pang, W. He, X. Jiang, and P. S. J. Russell, Nat.
Photonics 10, 454 (2016).
[6] P. J. Ackerman and I. I. Smalyukh, Phys. Rev. X 7, 011006
(2017).
[7] P. Grelu and N. Akhmediev, Nat. Photonics 6, 84 (2012).
[8] M. Stratmann, T. Pagel, and F. Mitschke, Phys. Rev. Lett.
95, 143902 (2005).
[9] A. Armaroli, C. Conti, and F. Biancalana, Optica 2, 497
(2015).
[10] T. Dauxois and M. Peyrard, Physics of Solitons (Cambridge
University, Cambridge, 2015).
[11] N. J. Zabusky and M. D. Kruskal, Phys. Rev. Lett. 15, 240
(1965).
[12] L. G. Wright, D. N. Christodoulides, and F. W. Wise,
Science 358, 94 (2017).
[13] J. M. Dudley, F. Dias, M. Erkintalo, and G. Genty, Nat.
Photonics 8, 755 (2014); G. I. Stegeman and M. Segev,
Science 286, 1518 (1999).
[14] S. Strogatz, Nature (London) 410, 268 (2001).
[15] J. Denschlag, J. Simsarian, D. Feder, C. W. Clark, L. Collins,
J. Cubizolles, L. Deng, E. W. Hagley, K. Helmerson, and
W. P. Reinhardt, Science 287, 97 (2000).
[16] A. F. J. Runge, N. G. R. Broderick, and M. Erkintalo, Optica
2, 36 (2015).
[17] P. Ryczkowski, M. Närhi, C. Billet, J. M. Merolla, G. Genty,
and J. M. Dudley, Nat. Photonics 12, 221 (2018).
[18] A. Hause and F. Mitschke, Phys. Rev. A 88, 063843 (2013);
K. Lakomy, R. Nath, and L. Santos, Phys. Rev. A 86,
013610 (2012).
[19] X. Liu, X. Han, and X. Yao, Sci. Rep. 6, 34414 (2016).
[20] A. Zavyalov, R. Iliew, O. Egorov, and F. Lederer, Phys. Rev.
A 80, 043829 (2009).
[21] Y. Wang, F. Leo, J. Fatome, M. Erkintalo, S. G. Murdoch,
and S. Coen, Optica 4, 855 (2017).
023905-6
[22] J. Herrmann, Opt. Commun. 98, 111 (1993).
[23] N. W. Pu, J. M. Shieh, Y. Lai, and C. L. Pan, Opt. Lett. 20,
163 (1995).
[24] H. Li, D. G. Ouzounov, and F. W. Wise, Opt. Lett. 35, 2403
(2010).
[25] K. Goda, K. K. Tsia, and B. Jalali, Nature (London) 458,
1145 (2009).
[26] Y. Xu, X. Wei, Z. Ren, K. K. Y. Wong, and K. K. Tsia, Sci.
Rep. 6, 27937 (2016).
[27] A. Mahjoubfar, D. V. Churkin, S. Barland, N. Broderick,
S. K. Turitsyn, and B. Jalali, Nat. Photonics 11, 341
(2017).
[28] See Supplemental Material at http://link.aps.org/
supplemental/10.1103/PhysRevLett.121.023905 for more
detailed discussions.
[29] See Supplemental Material at http://link.aps.org/
supplemental/10.1103/PhysRevLett.121.023905 for a
movie showing the experimental real-time measurements
from direct measurement (i.e., not using TS DFT) for the
formation and evolution of soliton molecules.
[30] See Supplemental Material at http://link.aps.org/
supplemental/10.1103/PhysRevLett.121.023905 for a
movie showing the experimental real-time measurements
from DCF (i.e., with the TS DFT technique) for the
formation and evolution of soliton molecules.
[31] R. Dunsmuir, J. Electron. Control 10, 453 (1961).
[32] I. Mozjerin, S. Ruschin, and A. A. Hardy, IEEE J. Quantum
Electron. 46, 158 (2010).
[33] S. Gao, C. Yang, X. Xiao, Y. Tian, Z. You, and G. Jin, Opt.
Express 14, 2873 (2006).
[34] X. Liu, Phys. Rev. A 77, 043818 (2008).
[35] S. M. J. Kelly, Electron. Lett. 28, 806 (1992).
[36] X. Liu, Y. Cui, D. Han, X. Yao, and Z. Sun, Sci. Rep. 5,
9101 (2015).
[37] H. A. Haus and W. S. Wong, Rev. Mod. Phys. 68, 423
(1996).
[38] K. Goda and B. Jalali, Nat. Photonics 7, 102 (2013).
[39] K. Goda, D. R. Solli, K. K. Tsia, and B. Jalali, Phys. Rev. A
80, 043821 (2009).
[40] X. Wei, B. Li, Y. Yu, C. Zhang, K. K. Tsia, and K. K. Y.
Wong, Opt. Express 25, 29098 (2017).
[41] A. G. Rozhin, Y. Sakakibara, and S. Namiki, Appl. Phys.
Lett. 88, 051118 (2006).