Ultrafast laser spectroscopy uncovers mechanisms of light energy conversion in photosynthesis and sustainable energy materials

With their invention, ultrashort pulse lasers have opened a wide range of new possibilities in chemical physics research and continue playing an important role. Many discoveries would not have been possible without the direct access to the timescale on which processes involving energy, electrons, atoms, and molecules occur. “Ultrashort” is of course a somewhat arbitrary concept and historically has often been used to indicate the transition from the picosecond to the sub-picosecond or femtosecond timescale. In this review, we have, however, chosen to include both the picosecond and the femtosecond timescales or, to define it more precisely, experimental work performed with both picosecond and femtosecond pulses. The ensuing dynamics following initiation with a picosecond or femtosecond pulse may start on the shortest femtosecond timescale, but carry on to much longer times, ps, ns, μs, and ms. With this definition of ultrashort pulses, our account includes the time from 1966, the year of the discovery of passive mode-locking in Nd-glass lasers and generation of the first ps pulses,¹ up to the date of writing (self-locking of laser modes was in fact first demonstrated in a Q-switched ruby laser, but generated ∼1 ns pulses).² Very recently, the generation of sub-femtosecond pulses (down to a few tens of attoseconds) has become possible and a new field, “attochemistry” is now emerging, dealing with the most fundamental atomic and molecular processes (for a review see, e.g., Ref. 3). This kind of ultrafast work is not discussed in this review.

Lasers in chemical physics research is an immense field, impossible to cover in a single review. We have, therefore, decided to first narrow the scope of our review to studies employing ultrashort pulse lasers. This is also a very large field, including areas like chemical reaction dynamics, general photophysics and photochemistry of molecular systems, energy- and electron transfer (ET), etc., showing the impact of time-resolved spectroscopy studies on chemical physics research. Therefore, we have further narrowed the review to research areas where we ourselves have been/are active, fields which we believe interest many colleagues and scientists, and which are currently also of general interest. The impact of ultrafast studies on our understanding of chemical reaction dynamics is also nicely illustrated by the Nobel Prize in chemistry 1999 awarded to Ahmed H. Zewail.^4,5

The review will cover two linked areas, photosynthesis and sustainable energy materials, from the point of view of fundamental ultrafast processes. This means that we will discuss light-induced energy and ET, and related processes, in these systems. Photosynthesis research has a long history and already at an early stage of ultrafast spectroscopy, picosecond pulses were employed to study the primary ET processes in photosynthesis.^6,7 It is, therefore, perhaps not surprising that photosynthetic systems and the processes therein have inspired design and study of artificial systems for harvesting and utilization of solar energy. A vast number of papers dealing with ultrafast processes in photosynthetic systems, energy materials, and devices have been published. Therefore, we will not discuss every ultrafast work on a photosynthetic preparation, or a new solar cell material. Instead, we will try to identify major steps of progress within the chosen scientific areas and discuss the results that led to new insights and understanding of function and structure.

We will start this review with a short account of the development of generation of ultrashort pulses and a few of the most common measurement techniques. This will take us from the discovery of passive mode-locking of a Nd-glass laser¹ over passively and actively mode-locked dye lasers⁸ to Kerr lens mode-locking of the Ti:Sapphire laser. Stretching-amplification-compression of optical femtosecond pulses leading to the Nobel Prize awarded concept of chirped pulse amplification⁹ is a very important step forward leading to widely wavelength tunable femtosecond pulses. The x-ray free electron laser (XFEL) is the final (at this time) addition to the arsenal of ultrashort pulse lasers. Thus, the technical development over the last 30 years has taken us from ∼10 ps pulses at a few different¹⁰ wavelengths, typically 1060, 530, 355, and 266 nm, to pulses of variable duration covering the wavelengths from millimeters [terahertz (THz) pulses] to fractions of nanometers (hard x-rays). With these pulses, we are well equipped to study virtually any dynamic process in any atomic, molecular, or solid state system. We note that generation and use of attosecond pulses will not be discussed in our review, since the results of such studies still are of limited bearing on widely different chemical systems.

We will start the account of work using ultrashort pulses with studies of the primary light-induced¹¹ processes in photosynthesis. This includes the ET processes in reaction centers, energy transfer in photosynthetic antennas, as well as photophysics and photochemistry of photosynthetic chromophores. The photosynthetic light-converting machinery consists of extensive coupled pigment systems, including several types of chromophores. To monitor the flow of energy and charge through such a system challenges most aspects of a spectroscopic measurement. Very high signal-to-noise is required in combination with very low light intensity, high temporal resolution and wide wavelength tunability. We will show how the technical progress and development in ultrafast methods have made this feasible. Ultrafast studies of carotenoid photophysics and dynamics are an example of new findings enabled by ultrashort pulses. Much of the fascinating properties of carotenoids had escaped discovery by using more conventional stationary spectroscopic methods, but application of ultrashort pulses led to major new insights into carotenoid energetics and excited state dynamics. Triggered by new experimental possibilities afforded by multidimensional spectroscopy, many photosynthetic processes and mechanisms have been reexamined recently, and also a heated debate on the role of coherent energy transfer in photosynthesis. We will summarize the various results and describe the most recent findings.

A vast number of materials for sustainable energy production have been developed over the years and studied using ultrashort pulses. Here, we will discuss the results for a few materials for capture of solar energy and conversion to either electricity in solar cells or fuel through photocatalysis. We will begin this section with a short discussion of photophysics and photochemistry of transition metal complexes since they are finding potential use both as sensitizers in solar cells and as photocatalysts. This includes excited state dynamics as it appears in the most used complexes, Ru-based and similar, as well as very recently developed Fe-based complexes with greatly extended charge-transfer (CT) state lifetimes, challenging traditional Ru-complexes as the obvious choice for photofunctional applications. We will describe the role of ultrafast studies in characterizing and understanding the excited state properties of these remarkable molecules.

Transition metal complexes have also played an important role in the development and implementation of ultrafast x-ray spectroscopy and scattering. Due to the heavy atoms with many electrons, they generate strong signals and have therefore been somewhat of a test ground for the introduction of these methods at synchrotrons and XFELs. Once optimized, the return has been rewarding and the species and state selectivity of x-ray methods have delivered new knowledge on for instance relaxation pathways and state characteristics, not obtainable with optical methods. We will show how this was possible.

The dye-sensitized solar cell (DSC), with an active material consisting of a nanocrystalline metal oxide thin film sensitized to visible light by a dye, was invented in 1991 by Grätzel and O'Regan.¹² Since then, thousands of dye–semiconductor combinations with various electrolyte systems have been developed and led to a slow but steady improvement of light to charge conversion efficiency, from ∼7% to today's ∼15%.¹³ We will discuss light-induced dye-to-semiconductor electron injection, recombination as well as charge carrier mobility and transport, key processes for understanding the function of a DSC that start on the ultrafast timescale and may proceed into the nano-, micro-, and millisecond time domains.

The so-called bulk heterojunction (BHJ) concept (see, e.g., Ref. 14) is the most successful design of a fully organic solar cell material. From a superficial view, it somewhat resembles the organization of a DSC material; it has a light-absorbing material, a conjugated polymer, acting as electron donor and a fullerene or another polymer as electron acceptor, all blended into nanodomains, or down to the molecular level. Also here, charge generation, recombination, and transport are the fundamental processes underlying the function of a solar cell. DSC and BHJ materials with their nanostructured light-harvesting and electron donor/acceptor organization have often been compared to photosynthetic systems and inspiration for their design borrowed from the latter.^15,16

As we will see from our presentation below of DSC and BHJ dynamics, decisive functional steps occur on the ultrafast, often sub-100 fs, timescale, frequently in competition with other less productive relaxation processes. Ultrafast methods are, therefore, powerful tools to monitor reaction pathways and identify bottlenecks toward optimizing light-to-charge conversion efficiency of the materials. Our review will illustrate this with several examples.

Organometal halide perovskite (OMHP) materials are the most recent addition to the family of photovoltaic (and luminescent!) materials, which in just a few years have risen to a serious competitor to more established inorganic semiconductor materials. The interplay between electronic and structural degrees of freedom in OMHPs results in unusually rich and interesting dynamics, which the wide variety of ultrafast tools is very well suited to disentangle. We will discuss the results illustrating the complex carrier and structural dynamics of the perovskites, providing some answers to their remarkable properties.

Solar energy, like wind energy, is intermittent and its optimal utilization therefore requires storage possibilities. This can be, for example, achieved with solar cell driven electrolysis of water, but a more direct way of achieving it would be through photocatalytic processes emulating photosynthetic water oxidation and production of molecular hydrogen, or some other fuel. Development of photocatalysts and studies of the light-induced processes have recently appeared as a new research field on its own. Here, we will summarize some of the ultrafast spectroscopy work aimed at characterizing initial photoinduced energy- and ET processes in a few model catalysts.

This is not a review of the individual photosynthetic systems or materials designed for harnessing solar energy by conversion to electrons (i.e., solar cell materials) or energy-rich molecules (e.g., solar fuel). Instead, the main aim with this review is to illustrate how ultrafast spectroscopy, from the early days of its conception until today, has contributed to our understanding of the function of nature's most important process, photosynthesis, and novel solar energy materials. In our discussion of the various photosynthetic and material systems, we have, therefore, focused on the achievements, and future perspectives will be only briefly discussed when ultrafast spectroscopy clearly could shed new light on unresolved issues. In any molecular or solid state material, in general, whose function emanates from interaction with electromagnetic radiation by absorption of photons, the functional processes compete with radiative or radiationless decay of excited states. To be efficient, the functional processes have to outcompete these “non-productive” processes that often proceed on the fs to ns timescale. This means that in order to get insights into function for any such system, it is necessary to study and understand the various processes that start in the excited state and eventually lead to a product or some nonproductive state (e.g., the ground state). Thus, ultrafast spectroscopy, in its many incarnations, has been used to disentangle the processes leading to all possible final states.

The ultimate timescale of processes involving the smallest chemical entities, electrons, atoms, and molecules, is set to femtoseconds and picoseconds by the smallness of these units and their characteristic motions. The direct access to this, the atomic and molecular timescale, has been approached through a stepwise technical development for generation of short pulses of electromagnetic radiation (light from now on), over a period of 70 years or more. Strategies relying on the generation of short bursts of other form of energy, i.e., electrons^17,18 or heat,¹⁹ or high frequency modulation of continuous light,²⁰ have also been developed, but the concept involving light pulses has turned out to be the winning approach with detection techniques relying on the principle of photography.

This development started with the introduction of the flash photolysis technique using millisecond flashlamp pulses in the late 1940s and early 1950s.²¹ The flash of light initiated a chemical reaction in the studied sample and the progress of the process was monitored by measuring changes in absorption of a continuous light beam detected with a photomultiplier and time resolved with for instance an oscilloscope. Much later with the use of increasingly shorter light pulses, the technique has been named pump-and-probe and many different spectroscopic techniques have been used for detection. In such an experiment, both pump and probe are short pulses of light, and time resolution is achieved by a variable time delay between pump and probe, realized with an optical delay line; the probe pulse simply travels a variably longer distance than the pump pulse. This means that no fast detection equipment is needed—an optical path length of 0.3 mm corresponds to 1 ps and time delays down to sub-fs times can easily be achieved. Early versions of pump-probe measurements relied on lasers with low pulse repetition rate, or even single pulse lasers. Modern pulsed lasers have kHz to MHz pulse repetition rates, which implies that a measurement in effect is repeated at this frequency leading to powerful signal averaging. Following this section on short pulse generation, we will describe a few of the most used detection techniques.

With the advent of the laser and giant pulse formation, or Q-switching, in the early 1960s^22,23 the microsecond and nanosecond timescales were opened for investigation. Only a few years later, in 1966 passive mode-locking of a solid state Nd³⁺-glass laser, with the help of a saturable absorber dye, produced ∼10 ps pulses;²⁴ similar mode-locking of the related Nd³⁺-YAG laser, having somewhat narrower spectral bandwidth of the active laser medium, resulted in ∼30 ps pulses. This development paved the way to the new field of picosecond spectroscopy. These solid state lasers were pumped by flash lamps at low repetition rates, typically a few Hz or lower, and the resulting pulses were of high energy. This often caused unwanted non-linear effects in the studied samples, which led to distorted spectral and kinetic response. The low repetition rate and large pulse-to-pulse intensity variation, in addition, made signal averaging of measured responses practically impossible.

Mode-locked dye lasers, which typically were pumped by a continuous or mode-locked (MHz pulse repetition rate) ion laser produced intensity-stable low-energy pulses at MHz repetition rates. This made lock-in detection possible, which drastically improved the sensitivity of measurements. Pumped by a CW ion laser, the dye laser was passively mode-locked with a saturable dye solution.^25,26 In a linear cavity configuration pulses as short as 0.3 ps were reported,^27,28 whereas colliding pulse mode-locking in a circular cavity configuration generated much shorter, less than 100 fs pulses,²⁹ and with dispersion compensation pulses as short as 27 fs could be generated.³⁰ The necessity to match absorption properties of lasing and mode-locking dyes for optimum pulse width limits the wavelength tunability to typically 10–20 nm; wavelengths within a range of a few tens of nanometers around 600 nm were therefore the most used.

Active mode-locking, or synchronously pumping with high repetition rate (typically ∼80 MHz) mode-locked ∼100 ps ion laser pulses, was the remedy to the limited tuning range. Virtually all laser dyes absorbing the ion laser wavelength could be used with synchronous pumping, resulting in ps pulses of a few ps duration from the near infrared to the green colors in the visible.⁸ Synchronous pumping of the dye laser did, however, not generate short sub-picosecond pulses as achieved with passive mode-locking.

The oscillator outputs of high repetition rate (typically ∼80 MHz) nanojoule pulses were often used directly in experiments, and sometimes at somewhat reduced repetition rates by Bragg cell cavity dumping or Pockels cell pulse picking. The high pulse-to-pulse amplitude and pulse width stability enabled lock-in amplifier detection and high signal-to-noise ratio measurements. In contrast to the early low repetition rate, high pulse energy, picosecond lasers this now allowed efficient signal averaging and measurements at very low excitation pulse energy, avoiding unwanted non-linear effects that distorted the response of the system under investigation. Measurements with these high rep-rate, low pulse energy lasers were generally one-color measurements, i.e., pump and probe were of the same wavelength. Dual-color pump-probe measurements with limited choices of pump and probe wavelength combinations could, however, be realized with multiple dye laser oscillators, pumped by the same pump laser. Since both laser oscillators were pumped by the same laser, pulses from the two dye lasers were perfectly synchronized.³¹ 

The invention of the Ti:sapphire laser in 1982 (Ref. 32) and its Kerr-lens mode-locking³³ combined with chirped pulse amplification⁹ implied nothing less than a revolution for ultrafast science. A Ti:sapphire oscillator typically produced ∼650–1000 nm sub-100-fs transform-limited pulses, which could be stretched to durations such that amplification to high power still was below damage threshold of the amplifier material. With the help of prisms or gratings, the long amplified pulses could then be compressed back to the original transform-limited duration. The peak power of the amplified and compressed pulses is sufficiently high to facilitate frequency shifting over wide ranges with the help of a range of non-linear optical effects, e.g., harmonic generation (e.g., second-, third-, and fourth-harmonic), optical parametric generation and amplification (OPG/OPA), and white light continuum generation with high conversion efficiency. The result is sub-100-fs pulses of almost any wavelength from the far infrared to the ultraviolet that allow for studies of most atomic, molecular and material systems. Single-cycle THz³⁴ pulses of ps duration, generated through, for instance, optical rectification³⁵ and sub-fs soft x-ray pulses generated by high harmonics generation^36,37 extend this range further.

Synchrotrons and hard x-ray pulses opened this energy range of the electromagnetic spectrum for analysis in time-resolved studies, albeit with a time resolution limited by the ∼100 ps synchrotron pulses.^38,39 Pulse slicing⁴⁰ extracted a shorter temporal slice of the synchrotron pulse and enabled in principle measurements with x-ray probing with few-ps resolution. The low number of x-ray photons in an x-ray slice and the required very long measurement times, however, prevented the widespread use of this technique.

With the advent of the XFEL, this has all changed. With several facilities in operation worldwide, ultrafast x-ray spectroscopy and scattering (see, e.g., Ref. 41) measurements with sub-ps temporal resolution and probing over a wide range of x-ray energies are now becoming routine. The very short, often ∼10 fs, duration of the x-ray pulse implies that it is no longer the x-ray pulse itself that sets the limits for time resolution, but rather the duration of the optical laser pulse used for excitation and the synchronization between optical and x-ray pulses. Increasingly more precise timing tools to compensate for the jitter between optical and x-ray pulses pushes time resolution of experiments to the limits defined by the cross correlation between exciting and probing pulses.⁴² 

Dynamic chemical responses can be monitored and time resolved with the help of many different methods, absorption or emission of light, elastic or inelastic scattering, non-linear responses, etc. Absorption or emission of light is perhaps the conceptually most straightforward methods and depending on which part of the electromagnetic spectrum is used, they form a variety of different spectroscopies—starting from the low energy end of the spectrum, e.g., THz spectroscopy, IR spectroscopy, uv/vis spectroscopy, x-ray spectroscopy. In all methods where detection is realized with the help of a probe beam (e.g., absorption and scattering processes), time resolving the sample response relies on the pump-probe concept and a time delay between pump and probe pulses. Like all pump-probe techniques, transient absorption is a third-order nonlinear technique, meaning that during the measurement the sample interacts with three electric field instances. Time resolving of sample luminescence response can be achieved both electronically (e.g., with a streak camera) and based on the pump-probe concept in a gating experiment (e.g., fluorescence upconversion or Kerr gate detection). Detection in the early picosecond experiments was performed by monitoring changes in absorption (or transmission) of visible probe light, or time resolving sample luminescence, and still today these are extensively used methods. Below, we will describe these “standard” methods, as well as a few somewhat more elaborate methods.

To illustrate both the principle of a transient absorption measurement and the impressive improvements in measurement quality and accuracy enabled by technical progress over several decades, we show how a typical measurement was performed when one of us had the privilege to work as a Ph.D. student at Bell Laboratories in 1975. The experiment we have chosen was to characterize the intersystem crossing from the S₁ to the T₁ state in an aromatic molecule, acridine, by time resolving the process and to measure the absorption spectrum of the T₁ state.⁴³ 

The layout of the laser and detection system used for these measurements is illustrated in Fig. 1. The laser was a passively mode-locked Nd³⁺-glass laser amplified in two stages. The oscillator generated a microsecond-long pulse train of approximately hundred 10 ps pulses spaced by ∼10 ns, but only pulses in the beginning of the train were of sufficient quality to be used in the measurements. One single ps pulse was, therefore, extracted from the train by a Pockels' cell triggered by the high-voltage pulse generated by a laser initiated breakdown in a “spark gap.” The pulse repetition rate of the laser was one “shot” every ∼10 min—or, every time the flashlamp and capacitor bank used for firing the lamp discharge had recovered from the previous shot. This means that a measurement of a kinetic trace or a spectrum had to be completed with one single pulse, since the low reproducibility of pulses made averaging meaningless. How do you do that? Probe light was, like often still today, a white light continuum, but with one big difference. White light continuum generated with picosecond pulses of insufficient peak power to saturate the non-linear effects is of poor quality, with large intensity and spectral variations from pulse to pulse. “Scanning” time delay between pump and probe pulses was performed by sending the white light continuum through an “echelon” (item 11 in Fig. 1)—a stair-like piece of quartz breaking up the continuum pulse into several segments separated in time by the difference in thickness of the quartz stairs. As illustrated in Fig. 2, ten time points or so could be obtained covering a time window of a few-100 ps. A sample and reference beam for measuring I and I₀ was absolutely necessary to compensate for the spatial intensity variation of the continuum after passing the echelon. Both I and I₀ beams were sent through a spectrometer and registered by an optical multichannel analyzer. Although no error bars are indicated in the kinetic trace of Fig. 2, the accuracy was probably no better than ± 0.1 ΔOD unit.

Since then, progress in ultrashort pulse generation and detection technology over several decades has made transient absorption measurements a versatile and powerful technique. Key elements of this development are femtosecond pulses of almost any wavelength at high and variable repetition rates with high pulse-to-pulse intensity and pulse width stability. This enables studies of a broad variety of atoms, molecules, materials, and processes with detection of very weak signals with ΔODs on the order of 10⁻⁵–10⁻⁶.^44,45 Complete transient absorption spectroscopy systems are available for purchase since a couple of decades, which further spreads the use of the technique.

Transient THz spectroscopy

Transient absorption spectroscopy is generally performed with probe pulses in the IR-UV spectral regions, but such probes are not always best suited for unraveling photoexcited dynamics. For example, the function of, e.g., photovoltaic materials rely on the generation of mobile charges that can be harvested as a photocurrent (PC). Conventional transient absorption spectroscopy is insensitive to the dynamics (mobility) of free charges. However, to directly access the behavior of charge carriers on the ultrafast timescale, time-resolved terahertz (THz) spectroscopy (TRTS) is a powerful tool. Upon pulsed (∼80–100 fs) light excitation with a photon energy above the bandgap of a molecule or solid state material, charged species, either loosely or tightly bound, are generated. This results in a change of photoconductivity (Δσ) in the material, which will modulate the characteristics of pulsed THz radiation used as a probe. This change in Δσ can be calculated using the following relation:⁴⁶ 

where n_exc is charge density, e₀ is the elementary charge, φ is photon-to-charge conversion quantum yield, μ_e and μ_h are the electron and hole mobility, respectively, ΔE_exc and E_gs are the THz electric field transmitted through the sample with and without light excitation, respectively, ε₀ is permittivity of vacuum, c is velocity of light, F is the fluence of the excitation light in ph/cm², α is the absorption coefficient, and L—the thickness of the sample. The quantity that can be obtained from this equation is the total carrier mobility in cm²/V s. As shown, the measured change in photoconductivity, Δσ, is a product of quantum yield and mobility, meaning that in order to obtain Δσ, the photogenerated species should be both charged and mobile. A tightly bound molecular exciton, which may be created by light excitation, will not be detected since by definition it has no net charge. In the same manner, if the pump pulse creates ions, whose mobility is very low, the sensitivity of a particular measurement may not be sufficient to detect them. The temporal evolution of the charge population and mobility defines the shape of the THz transient photoconductivity kinetics. On the one hand, a rise in photoconductivity kinetics reflects generation of charged species and/or increase in mobility of the charges. On the other hand, a decay represents the decrease in the mobility (e.g., due to relaxation or trapping) and/or disappearance of charge carriers either by recombination, or injection to a low-mobility acceptor material.

Pulsed THz radiation is conveniently generated through an optical rectification process by pumping a ZnTe crystal with 800 nm, ∼100 fs, and 100 μJ pulses.⁴⁷ Another ZnTe crystal is used for detection by spatially and temporally overlapping the pulsed THz radiation with 800 nm gating pulses in a process known as electro-optical sampling. The transient THz photoconductivity kinetics is collected by fixing the delay of the 800 nm gating pulses at the peak of the THz electric field and scanning the pump-probe delay within a desired time interval, typically up to 1 ns. To obtain a THz photoconductivity spectrum, the pump-probe delay is fixed at a desired pump-probe delay while the delay of the 800 nm gating pulses is swept to map the THz electric field. The measured Δσ as a function of THz frequency represents the photoconductivity spectrum, and from the shape of this spectrum carrier scattering time and carrier localization can be determined.^46,48 It should be noted that at the earliest timescale, φ is often assumed to be unity, while at longer times this represents the change in charge population at a particular time. φ close to 1 means that all absorbed photons are converted to mobile charges. However, since accurate measurement of φ is generally difficult, this assumption means that reported mobility values are lower limits and can in reality be considerably higher.

For time and spectrally resolving of sample luminescence four different methods are in frequent use, two based on fast electronic detection [time correlated single photon counting (TCSPC) and streak camera] and two relying on the pump-and-probe principle through a gating process (fluorescence Kerr gate and fluorescence upconversion). All four techniques are described and discussed in the excellent book by Lakowicz,²⁰ so only a very brief description is given here.

Both TCSPC²⁰ and streak camera measurements⁴⁹ are very well-established techniques with complete measurement equipment and devices available from several commercial vendors. TCSPC is designed to detect single fluorescence photons from a periodic signal excited by a pulsed laser with high repetition rate. A time-to-amplitude converter converts the short time period between the excitation laser pulse (start pulse) and the first detected fluorescence photon (the stop pulse) to a voltage, which is digitized in an analog-to-digital converter and stored in a multichannel analyzer. By repeating this many, many times a histogram of detected fluorescence photons at a particular arrival time representing the fluorescence decay is constructed. Since single photons are detected, this is a very sensitive method and the shortest decay times that can be retrieved with this method is 10–20 ps. TCSPC measures fluorescence decays at a single wavelength. If the time dependence of a fluorescence spectrum is needed, many time traces at different wavelengths have to be measured and combined into a spectrum.

A streak camera⁴⁹ is a device featuring two-dimensional detection, meaning that the full temporal evolution of a fluorescence spectrum can be directly measured. Time resolution in a streak camera is achieved by converting time into a distance on a detector phosphor. Fluorescence photons generate photoelectrons at a photocathode, and these electrons are deflected by a high-voltage sweep perpendicular to their direction of propagation in a streak tube. On the detector phosphor, they form a two-dimensional streak image where one axis represents time and the perpendicular axis—wavelength. Streak tubes with sensitivity from the x-rays to the infrared and featuring single photon detection can be obtained. Hamamatsu, one manufacturer of streak cameras, states that their single scan camera for single events has a time resolution of 100 fs, while their synchroscan camera for repetitive events has a time resolution of  ≤800 fs.

In both the Kerr gate⁵⁰ and upconversion^51,52 fluorescence detection time resolution is achieved with the help of an “optical shutter” that cuts temporal slices out of the fluorescence decay. The gate timing is controlled by a delay stage, which sets the time difference between fluorescence excitation pulse and gate pulse. Sweeping the delay and measuring the intensity of the fluorescence slices yield the decay profile. The Kerr gate consists of a liquid (e.g., CS₂) or solid material (e.g., thin glass slide) placed between two crossed polarizers. In the non-active state, the Kerr medium is isotropic and no fluorescence is transmitted through the gate. Applying an optical gate pulse induces birefringence in the Kerr medium, which rotates the polarization of transmitted fluorescence and causes a portion of the fluorescence to pass through the second polarizer. The duration of the light burst transmitted through the Kerr gate, and thus, the time resolution is determined by the duration of the gate pulse, the relaxation time of the induced birefringence and the thickness of the Kerr medium. In the original setup developed by Duguay,⁵⁰ the fluorescence spectrum was integrated, but since the Kerr gate has no wavelength discriminating components, it is in principle a broad-band method allowing broad spectrum detection. This was later implemented by Schmidt and co-workers in a modified setup.⁵³ With 100-fs gate pulses, a Kerr medium with short relaxation time can be chosen, such that time resolution is mainly limited by the duration of the gate pulse.

In a fluorescence upconversion experiment,^51,52 the gate medium is a non-linear crystal and the gating mechanism is sum frequency generation between the fluorescence light and the gating pulse. Other nonlinear frequency conversion modalities, such as difference frequency generation, can be also used for the gating. An upconverted signal is only present when both fluorescence and gate pulse overlap temporarily and spatially in the non-linear crystal and frequency upconverted radiation (ν_fluorescence + ν_gate) is generated. The time resolution of an upconversion experiment is mainly determined by the duration of the gate pulse and thickness of the non-linear crystal. As originally developed, fluorescence upconversion was a single-color technique—the phase matching angle of the non-linear crystal had to be optimized for each fluorescence wavelength.⁵¹ Later, a broadband version of the method was developed, with a judicious choice of non-linear crystal and phase matching conditions.⁵⁴ 

To obtain dynamical structural information about molecules and different materials, that is, ultrafast changes of structural configurations following light excitation, probing vibrational transitions after the excitation pulse is desirable, as different structural conformations give rise to different vibrational responses. There are several time-resolved methods, including femtosecond stimulated Raman spectroscopy (FSRS), pump-degenerate four wave mixing (DFWM), and visible-IR two-dimensional spectroscopy, etc., capable of such investigations. We start with the FSRS technique, which is a fifth-order (in interaction with the electromagnetic fields) nonlinear technique that probes vibrations on the excited electronic states.

Time-resolved Raman spectroscopy was first implemented in the nanosecond and then picosecond time domains. In 2000, Yoshizawa and Kurosawa used a trick of separating the narrowband Raman pump pulse and the broadband Raman probe pulse in FSRS to push time resolution into femtoseconds.⁵⁵ The typical implementation of FSRS involves three pulses: the first actinic pulse, usually in the UV or Vis range, is used to promote a substantial part of the sample to the excited state. Then after some delay time, simultaneously, the narrowband Raman pump pulse (1–4 ps long) and broadband Raman probe pulse (<30 fs) are sent, and changes in spectral amplitude of the probe pulse are detected.⁵⁶ Positive or negative Raman lines are detected on top of the smooth probe spectrum with the spectral resolution of 10 cm⁻¹. Since the free induction decay signal of the superposition of vibrational states is measured, which takes a picosecond or even longer to develop, the “real” resolution of the experiment is not femtosecond. However, one captures the system state at the moment of the arrival of the femtosecond probe—in a sense at the moment when free induction decay is starting—so typically it is said that, similar to pump-probe, the time resolution of the FSRS experiment depends on the length of the actinic and probe pulses. The signal is corrected for probe fluctuations and background. For this, the Raman probe spectral profile, but also the solvent Raman signals have to be measured and subtracted. Sometimes the ground state signal of the system is also subtracted to leave only the excited state contributions. Both non-resonant and resonant Raman pump pulses have been used in the experiments,^57–59 and the two approaches have their own advantages and disadvantages. For example, changing Raman line shapes were investigated, by tuning Raman pump frequency across the excited state resonance.⁵⁸ 

Another fifth-order non-linear technique sensitive to vibrational signals is pump-DFWM.^60,61 In contrast to FSRS, this is a fully time domain technique. The first step in the experiments is the same—the strong actinic pump pulse is used to excite the sample to an excited electronic state. This excitation is followed by a sequence of three pulses constituting the four wave mixing arrangement. The three pulses are pump, Stokes, and probe. The pump and Stokes pulses prepare a vibrational superposition (a wavepacket) in the excited state (vibrational “coherence” signal), which is then probed by the delayed probe pulse. The usually used homodyne detection of the oscillating signals requires some monotonously decaying signal, which is conveniently provided by the population dynamics signal. The vibrational spectrum in the excited state is obtained by Fourier transforming the signal dependence on the delay time between the pump/Stokes pulses and the probe. The time delay, corresponding to the delay in pump probe spectroscopy, is the delay between actinic and pump/Stokes pulses.⁶¹ 

Both FSRS and pump-DFWM provide complementary information to the more common pump-probe and time-resolved fluorescence techniques, because they directly probe vibrational transitions in the excited state. However, generally a large variety of signals can be measured in fifth-order techniques and great care has to be exercised to distinguish the signals of interest. Consequently, interpretation of the signals is often not straightforward.

Inspired by the tremendous success of multidimensional NMR spectroscopies, the analogous experiments were implemented in both the IR⁶² and visible wavelength ranges,^63,64 as well as the combination of the two.⁶⁵ Most commonly, the results obtained with these techniques are presented in two dimensional graphs, involving excitation and detection frequencies, and therefore, all nonlinear spectroscopy methods listed here are called two-dimensional spectroscopies. The main idea here is to measure correlations between initially excited transitions and transitions probed after the delay time.⁶⁶ The evolution of these correlations is then represented in the 2D maps. Just like pump-probe spectroscopy, 2D spectroscopy is a third order non-linear spectroscopy technique. This four-wave-mixing experiment relies on the assumption that the sample interacts with three weak exciting fields (pulses), whereupon transient polarization is created in the sample, which is then emitted as electromagnetic radiation and measured using heterodyne detection. Two main differences and at the same time advantages can be listed when compared to conventional pump-probe spectroscopy. First, full signal information, i.e., amplitude and phase, is measured in 2D spectroscopies, and second, excitation frequency resolution is achieved.

As 2D spectroscopies enjoy widespread popularity, multiple realizations in different spectral ranges can be found using various geometries, where phase matching conditions and/or pulse manipulation with pulse shapers are implemented. As a typical representation, here we focus on 2D electronic spectroscopy (2DES) in a non-collinear configuration,^64,67 where four pulses are used in the experiment. The first two pulses with a coherence delay introduced between them with interferometric accuracy, t₁, correspond to the single pump pulse in pump-probe spectroscopy. The second delay, t₂, between the second and the third pulse, called population delay, corresponds to the time delay in pump-probe experiments. Finally, a signal is emitted following the third pulse at a delay t₃. The fourth so-called local oscillator pulse is phase locked to and delayed in respect to the third pulse. This is necessary for heterodyne detection, where an interferogram between the signal and local oscillator is measured in a spectrometer. Usually, four pulses with wave vectors k₁ k₂, k₃, and k_LO are geometrically arranged in the corners of a square—facilitating measurements of only the phase matched signals in the k_S = −k₁ + k₂ + k₃ direction, which coincides with the direction of the local oscillator pulse k_LO. In this way, the number of signals detected in the experiment is highly reduced, in turn aiding interpretation of measured data.

The signal is extracted from measured interferograms by linear spectral interferometry techniques using Fourier filtering⁶⁸ and has information on both amplitude and phase. Thus, full information is obtained on the third order polarization created in the sample, in contrast to pump- probe measurements, where only amplitude is measured. The extracted signal has detection frequency information, whereas a Fourier transform over the coherence time completes the procedure and yields the excitation frequency information.

The spectral information that can be obtained about the system under investigation is limited by the laser spectrum used in the experiment. Therefore, spectrally broadband and ultrashort laser pulses are highly preferable, and thus, 10–15 fs pulses are often used. Consequently, the time resolution of the 2DES experiment is in the range of 20 fs. At the first glance, it seems paradoxical that high time and spectral resolutions are simultaneously achieved, both for excitation and detection frequencies. This could be interpreted as contradicting the time-frequency uncertainty principle regarding the laser pulses used in the experiment, which indeed limits spectral resolution of pump-probe experiments. This apparent paradox is overcome, because excitation frequency resolution is not achieved in a single experiment, but reconstructed with the help of a Fourier transform from multiple experiments where time delay between the first two pulses is scanned.

As already mentioned, the main advantage of 2DES is excitation frequency resolution, which simply allows one to “see the full picture” in a single experiment, whereas in more limited experiments some important details could be missed. Additionally, 2DES can further help disentangling several types of congested signals. This includes the possibility of separating rephasing and non-rephasing signals, which are inseparably mixed in pump-probe measurements. This separation is of great help when disentangling oscillating (or coherence) signals. Furthermore, phase information present in 2DES experiments allows for identifying the “direction” of phase evolution of coherence signals. An everyday analogy for this is the propagation direction of ripples caused by a stone thrown in a pond. In other words, whereas in 2DES these signals evolve on the full complex plane, only the real part projection is seen in pump-probe measurements. Finally, 2DES facilitates manipulation of all four “waves” in the four-wave-mixing experiment. By controlling polarization of the three interacting pulses and filtering the polarization component in the detected signal, a wide range of anisotropy-type of experiments can be carried out. An example of polarization-controlled 2DES experiment, beyond that available in pump probe, is (45°, −45°, 90°, 0°),^69–71 where the angles in parenthesis indicate the relative orientation of the linear pulse polarizations and filtered polarization of the signal. This polarization sequence, called double-crossed polarization, allows for suppressing all signals except coherences, which are excited via interaction with the transitions having non-parallel dipole moments. This implies that only purely electronic coherences, or vibrational coherences excited via vibronically mixed states are selected. This is arguably the only method, which allows for direct detection of vibronic mixing in molecules⁷² and molecular complexes.^71,73,74

All photosynthetic processes in nature occur under ambient light, while ultrafast spectroscopy makes use of laser pulses, which are qualitatively very different from sunlight. The laser light used in ultrafast spectroscopy is coherent, usually highly polarized and has a large number of photons packed into a very short duration in time. Laser light also has a frequency comb spectrum making it very different from sunlight experienced by photosynthetic organisms in their natural environment. Thus, one fundamental question is how the large qualitative difference between experimental (ultrashort laser pulses) and natural (sunlight) light sources may affect the interpretation of experimental results toward understanding of photosynthetic functions (and human-made solar energy conversion).

In photosynthesis, the light intensity is traditionally specified in micro-Einsteins per square meter per second (μE·m⁻²·s⁻¹) and the light intensity corresponding to sunlight vary between 200 and 2000 μE·m⁻²·s⁻¹.⁷⁵ Since the unit of Einstein denotes one mole (6.023 × 10²³) of photons, this number can be recalculated to the photon flux of approximately 10²⁰–10²¹ photons·m⁻²·s⁻¹. Then, for chlorophyll which has an absorption cross section of σ ∼ 1 Ų,⁷⁵ this translates to 1–10 absorbed photons per chlorophyll per second. Thus, a chlorophyll molecule even at full sunlight is promoted to its excited state approximately once per 100 ms. This is in vast contrast to photon excitation densities typically used in ultrafast spectroscopy. The excitation intensities used in a typical transient absorption experiment employing a 1 kHz pulse repetition rate are in the range of 10¹³–10¹⁴ photons cm⁻²·pulse⁻¹, which, with pulse durations of 30–100 fs, corresponds to a photon flux reaching values as high as 10³⁰ photons·m⁻²·s⁻¹ during the pulse duration, thus many orders of magnitude higher than full sunlight. Of course, since for 1 kHz repetition rate we have only 1000 such intense short pulses per second, the total number of photons hitting the sample within one second is comparable to that under full sunlight. However, the photon flux within the ultrashort pulse is enormous and incomparable with natural light conditions.

The key question in the field of ultrafast spectroscopy of photosynthetic, or other biological systems, is how these large photon fluxes may affect the outcome (and interpretation) of experiments. It is important to realize that if a chlorophyll molecule under full sunlight is excited approximately once per 0.1 s, it means that even for the largest antenna systems at a given time moment there is never more than one excitation per whole antenna (this holds also for other photosynthetic pigments, bilins, and carotenoids since their molar absorption coefficients are comparable). If we assume a model antenna system containing a connected array of 100 pigments [e.g., the photosystem I (PSI) core, see Sec. III E 2 a], it will be excited by sunlight with a rate of about one excitation per millisecond. Thus, if the trapping time of such a system is ∼100 ps, most of the time there is actually no excitation in the system as the trapping time is seven orders of magnitude faster than the rate by which the system is excited.

This situation, however, differs for photon fluxes used in ultrashort laser pulses. We again assume the same model antenna of 100 pigments, each having a molar absorption coefficient approximately equal to that of Chl a, which results in a total cross section of the whole antenna of σ ∼ 4 × 10⁻¹⁴ cm². Then, if the intensity of the laser pulse (I) is given in photons cm⁻²·pulse⁻¹, the average number of pigments excited simultaneously by such a pulse is $x=I⋅σ$⁠. Using Poisson distribution, we can calculate the probability of simultaneous excitation of n pigments within our model antenna,

Then, P₀ is the probability that no pigment is excited, P₁ gives the probability of excitation of one pigment in our model antenna and so on.⁷⁶ These probabilities and average numbers of simultaneously excited pigments in our model antenna are given in Table I for some typical excitation intensities used in ultrafast experiments.

The table shows that for such a model antenna, using intensities >10¹³ photons pulse⁻¹ cm⁻² inevitably leads to a non-negligible probability of exciting more than one pigment within the antenna. If these two pigments are connected via some energy transfer pathway and their excited state lifetime in the antenna is longer than the time it takes for the two pigments to meet, it will lead to annihilation resulting in excitation intensity dependent dynamics (Sec. III). It is extremely important to consider this effect especially for large antenna systems such as chlorosomes (Sec. III C 2), for which excitation intensities as low as ∼10¹¹ photons pulse⁻¹cm⁻² are required to obtain annihilation-free dynamics.⁷⁷ The presence of multiple excitations within a connected array of pigments may also lead to population of higher excited states that may initiate photophysics that is not related to the initially excited state, complicating interpretation of data.⁷⁸ 

While the problems related to high photon fluxes resulting in annihilation are well understood and can be taken into account in data analysis,^79,80 the issues related to the fundamental difference between incoherent sunlight and coherent ultrashort laser pulses are much more difficult to evaluate. The coherent vibrations induced in photosynthetic proteins after excitation by sub-100 fs pulses were reported already in 1991 in photosynthetic reaction centers⁸¹ and in 1994 for a light-harvesting antenna.⁸² However, the role of coherence in photosynthesis has become a highly debated topic more than a decade later, and 2DES experiments played a central role (Sec. II B 3 c). The first 2DES experiments on photosynthetic systems suggested the key role of coherences for the efficiency of energy transfer, leading to the advent of the field of “quantum biology.”⁸³ These and subsequent studies reporting the presence of coherences have led to a hypothesis that these quantum coherences are nature's tool to efficiently direct energy transfer through light-harvesting systems. Nearly fifteen years of discussions have eventually provided a better understanding of these phenomena, concluding that most of these long-lived coherences have their origin in collective vibrations generated by impulsive excitation by coherent ultrashort pulses.⁸⁴ A more detailed description of these phenomena is provided in Sec. III H.

Here, we would like to discuss an even more fundamental issue. All experiments addressing the ultrafast dynamics of photosynthetic systems are based on laser-induced coherent excitations, while under natural conditions photosynthetic systems are excited by incoherent sunlight. The relation between experimental observations and processes occurring in nature remains a matter of considerable debate as these two situations inevitably lead to a different response of a studied system.⁸⁵ The excitation with an ultrashort laser pulse represents a well-defined coherent field, having a coherence time much longer than the lifetime of the studied system, inducing a time-dependent perturbation. In contrast, sunlight is incoherent (the coherence time of just a few femtoseconds) and thus characterized as an ensemble of fields described by their statistical properties.⁸⁶ Thus, under incoherent sunlight the system is in a non-equilibrium steady state, free from any time-dependent coherences. Yet, static coherences, which may result from system-bath couplings, may occur in such a steady state regime.⁸⁷ In this non-equilibrium steady state, also the meaning of decay rates obtained from the time evolution of a system perturbed by an ultrashort pulse was suggested to be different.⁸⁵ 

Despite the concerns given above, there is no doubt that the experiments using ultrashort laser pulses, which are at the focus of this review, provide a wealth of information about the studied system. They give accurate information about the electronic structure, interaction of the pigments with their environment and, via measuring the decay rates, reveal the pathways and efficiencies of energy flow within the system. However, the exact description of how a light-harvesting system will behave under incoherent sunlight, which would correspond to a chlorophyll molecule absorbing approximately one photon per second, is still not completely clear. Theoretical approaches addressing the issue have appeared in recent years,⁸⁵ some of which also proposed how to assess the problem experimentally. It has been suggested that photon correlation spectroscopy, which relies on coincidence measurements of photons emitted by a system excited by incoherent, thermal light, can provide similar information as 2D spectroscopy.⁸⁸ Similarly, Hong–Ou–Mandel (HOM) interferometry measuring coincidences of entangled photons generated in a nonlinear crystal⁸⁹ may, if the studied system is placed into one arm of the HOM interferometer, provide information about the behavior of the light-system interaction in a single-photon regime corresponding to the conditions experienced under sunlight.⁹⁰ Other promising approaches using measurements of entangled photons to simulate sunlight conditions have been recently proposed,⁹¹ indicating that the coming decade may witness the dawn of experimental approaches that will be able to follow the response of a light-harvesting systems to incoherent sunlight by using quantum light.

Today, we know that the pigment system of photosynthetic organisms consists of two parts, one that collects the sunlight, called antenna, and another one, the reaction center (RC), that converts the transient excited state energy of the antenna to a long-lived electrochemical potential that drives all energy demanding processes of the organism.^79,92–94 Both antenna and RC are composed of pigment–protein complexes. An antenna, or light-harvesting complex (LHC), often consists of several polypeptides and tens to hundreds (or even thousands) of pigment molecules, while a RC generally is smaller with a few polypeptides and usually six pigment molecules. A functional antenna/RC assembly is often called a photosynthetic unit (PSU) and can consist of many LHCs per RC. The physical processes that achieve the light to electrochemical potential conversion is excitation energy transfer in the antenna and ET within the RC. Depending on factors such as involved pigments, distances, and the nature of interaction between pigment molecules, the timescale of this conversion ranges from femtoseconds to nanoseconds. The light-induced processes of photosynthesis are often studied in isolated LHC or RC preparations, but in order to characterize the energy flow through a functional antenna, or antenna-RC assembly, photosynthetic membrane preparations containing intact PSUs or whole organisms (i.e., algae, bacteria, and leaves) must be studied. The latter may pose a challenge in spectroscopic measurements because of a large number of spectrally overlapping pigments and strong light scattering that inevitably results in additional noise in the experiment.

Photosynthetic pigment–protein complexes, as well as membrane preparations have been studied using various methods of optical spectroscopy long before ultrafast measurements became possible. With the help of, e.g., fluorescence yield and polarized spectroscopy measurements, microscopic molecular level information was pursued (see Ref. 79 and references therein). Even before high-resolution crystal structures of RCs and light-harvesting complexes became available, simple geometrical models of pigment–protein complexes (see, e.g., Ref. 95) and lattice models⁷⁹ of whole PSUs or photosynthetic membranes were developed to retrieve this information.

Time-resolved measurements with temporal resolution capable of resolving the most elementary photophysical and chemical processes (<100 fs) were an important step toward identifying individual energy and ET steps, as well as energy and charge flow through extended pigment systems, but not until ultrafast dynamics were reconciled with high-resolution structures could pathways of energy and charge transport be identified with certainty (see, e.g., Refs. 96 and 97).

In order to illustrate the principles of light harvesting by photosynthetic antennas and energy conversion by RCs and to show how ultrafast spectroscopy has contributed to the understanding of the photosynthetic processes, we have chosen to discuss results for four main groups of photosynthetic organisms, cyanobacteria and red algae, green bacteria, purple bacteria and higher plants. This includes results for a number of isolated light-harvesting complexes, the two types of RCs, as well as results for photosynthetic membranes and intact organisms. The ET processes in the RCs are best studied in isolated RCs, because of often overlapping spectral properties of antenna and RC pigments, and since antenna to RC energy transfer is often understood to be a rate limiting process.

Photosynthetic organisms thrive under widely varying light conditions, from bright sunlight to a very dim glow of black smokers several kilometers down in the ocean. Our choice of organisms will illustrate nature's strategies to optimize the effectiveness of a particular organism under the conditions defined by intensity and spectral properties of light in its environment. On the individual pigment–protein complex level, our choice of organisms translates into the following LHCs: phycobilisomes of cyanobacteria and red algae, chlorosomes, the Fenna–Matthews–Olson (FMO) complex and B808–866 complex of green bacteria, LH1 and LH2 of purple bacteria, and the antenna complexes associated with photosystem I (PSI) and photosystem II (PSII) of plants. In all organisms, RCs are of either type I or type II and we will discuss results for both. Before we proceed with discussions of energy and ET processes in the photosynthetic pigment machinery, we make a short detour for considerations of the experimental conditions required for obtaining relevant information on light-driven energy and electron dynamics in photosynthetic pigment–protein complexes.

Already in the early days of ultrafast spectroscopy, the primary light-driven processes of photosynthesis were targets of several studies. It started with three studies of the charge separation (CS) in RCs of purple bacteria^6,7,98 and it was reported that the primary and secondary charge separation steps occur in less than 6 and 200 ps, values that by now have been verified over and over again after hundreds of more detailed and precise measurements. Studies of the energy transfer processes in photosynthetic antennas turned out to be more difficult. Fluorescence lifetimes of antenna complexes or photosynthetic membranes were reported to be excitation pulse intensity dependent,^99–101 and it was soon realized that the reason for this can be found in the fact that two or more excitations in a pigment system containing many pigment molecules coupled through energy transfer will extinguish each other until only one remains. The process called exciton–exciton annihilation is mentioned in Sec. II C, and there are two variants of it—singlet–singlet and singlet–triplet annihilation. Thus, if several pigment molecules are excited in an antenna system with coupled chromophores two excitations (excitons) will eventually meet on the same molecule (or a coupled oligomer) and generate a high-lying excited state. Such a state generally has a very short lifetime, limited by radiationless decay. The whole process results in a fast and non-exponential decay of the excited state population. This decay is solely a consequence of the simultaneous multiple excitation of the pigment system and obviously does not reflect the intrinsic decay properties of the population, because, as mentioned above, under natural conditions multiple excitation is not feasible. With the early low-repetition-rate high-energy (many photons) ps pulses, it was impossible to avoid these non-linear effects and obtain kinetics free of distorting exciton–exciton annihilation. Theoretical methods were, however, developed to benefit from the exciton–exciton annihilation, and the size of the pigment system, i.e., number of connected chromophores, could be estimated from the shape of a curve describing the dependence of integrated fluorescence yield on excitation pulse intensity.^102–104 Other parameters that can be estimated from the annihilation curve are the nearest neighbor energy transfer rate and energy trapping time by the RC. Thus, although these properties could not be obtained from direct kinetic measurements, they could be estimated from the annihilation behavior. Despite a model dependence built into the analysis that obviously has limited accuracy of obtainable information, many photosynthetic systems were studied, e.g., chloroplasts,⁹⁹ photosynthetic bacteria,¹⁰⁵ and phycobilisomes^106,107 and useful information was extracted. Excellent reviews of the annihilation phenomenon, theoretical models and experimental studies are given in Refs. 79 and 108. Distorting exciton annihilation effects on energy transfer kinetics of large pigment systems can obviously be eliminated by using excitation pulses of sufficiently low energy, such that only one pigment molecule per domain is excited.

In this section we start with a short discussion of excited state properties of photosynthetic pigments: chlorophylls, phycobilins, and carotenoids and proceed to discuss specific results for the various photosynthetic systems outlined above. We follow the same order of events as found in photosynthesis—light harvesting first followed by the charge separation in RC.

Knowledge of excited state properties of isolated pigments which play a role in photosynthesis is the key prerequisite for understanding the light-driven processes in photosynthesis. There are three types of pigments involved in photosynthesis: (1) chlorophyllides that are present in both LHCs and RCs in all photosynthetic organisms; (2) carotenoids, which are also found in the majority of LHCs and RCs, but there exist LHCs, such as FMO, without carotenoids; and (3) phycobilins that occur exclusively in phycobilisomes of cyanobacteria, red algae, and glaucophytes, described in Sec. III B. The three groups of pigments also differ in their diversity in photosynthetic organisms. Only four phycobilins are found in cyanobacteria and red algae, while 13 different chlorophyllides (six chlorophylls and seven bacteriochlorophylls) plus a few pheophytins, which are bacterio(chlorophylls) missing Mg atom, have been identified so far in various photosynthetic systems.⁷⁵ The pigment diversity is by far the largest among carotenoids. Even though only five carotenoid species are found in plants, more than hundred different carotenoids have been reported in various LHCs of photosynthetic micro-organisms.¹⁰⁹ 

Among the various photosynthetic pigments found in nature, the chlorophyllides are the most abundant. They include chlorophylls, such as chlorophyll a and b, (Chl a, b) and pheophytins found in photosynthetic systems of higher plants, as well as bacteriochlorophylls, such as bacteriochlorophyll a and c (BChl a, c) and bacteriopheophytins found in some species of photosynthetic bacteria. The functions of chlorophyllides include energy transfer and charge separation, and thus they serve as principle cofactors for the primary functions of photosynthesis. Currently, six Chls and seven BChls have been identified in various photosynthetic organisms (see Fig. 3 for the structure of Chl a). Already in 1881 Engelmann detected the action spectrum of photosynthesis in filamentous green algae, which closely resembled the absorption spectra of the Chl a and Chl b molecules. The latest discovered chlorophyll, Chl f, was reported in 2010, and was found in cyanobacteria from a stromatolite colony.¹¹⁰ The absorption spectrum of all Chls consists of two main spectral bands: the Soret band peaking in the 350–480 nm spectral range, and the lowest energy Q_y band extending from 630 to 720 nm for Chls and 700–1010 nm for BChls.⁷⁵ It is important to notice that the properties of the lowest energy Q band states (Q_y and Q_x) of chlorophyllides control the functions of these molecules. In the early sixties, M. Gouterman provided the explanation of the electronic structure of porphyrin type molecules.¹¹¹ In the Gouterman model, two independent electronic transitions named Q_x and Q_y with perpendicular transition dipole moments were identified. Later, with the help of polarization spectroscopy studies, it was shown that the angle between the Q transitions is closer to 70°.¹¹² In the absorption spectrum, transitions to these electronic states are accompanied by vibronic transitions involving coupled vibrational modes. Through the years, chlorophylls have been subject to numerous spectroscopic studies, and it thus comes as a surprise that the position of the higher-energy transition Q_x is still debated. In a seminal paper in 2013 Reimers and co-workers showed that available steady-state spectroscopic data, including absorption, emission, fluorescence excitation, linear dichroism, and magnetic circular dichroism, of 32 chlorophyllides can be fit with a relatively simple vibronic coupling model, involving one vibrational mode and strong coupling between Q_y and Q_x.¹¹³ Thus, it was suggested that in all of these molecules, the Q_y and Q_x states (transitions) are inseparably vibronically mixed. This means that the direction of transition dipole moments of the mixed states is also something in between x and y, in other words x-polarized absorption spreads through the whole Q band. However, finding an unambiguous experimental proof for this conjecture turned out to be difficult.

In addition to the singlet states discussed above, (B)Chls have lower lying triplet states, which can be populated via an intersystem crossing pathway from the Q_y state. The yield of this process was measured a long time ago and was concluded to be in the 50%–60% range for (B)Chl a and (B)Chl b (see Ref. 114) but recent data employing femtosecond transient absorption spectroscopy suggest lower yields around 30%.¹¹⁵ Triplet states do not directly participate in photosynthetic functions, but (B)Chl triplets can sensitize dioxygen producing highly reactive singlet oxygen, which can damage the photosynthetic apparatus. In vivo, both (B)Chl triplets and singlet oxygen are deactivated by carotenoid molecules (see Sec. III A 2).

A range of time-resolved spectroscopy studies have been performed to measure chlorophyll excited state lifetimes (see, for example, Ref. 116). Typical lifetimes of the singlet excited state (Q_y) are in the range of a few nanoseconds, whereas triplet states typically decay with a lifetime from tens of microseconds to a millisecond. The Q_y lifetime is readily measured by time-resolved fluorescence, and the first values were measured more than 60 years ago, providing lifetimes of 5.1 and 3.9 ns for Chl a and Chl b, respectively.¹¹⁷ Since then, a number of reports provided the Q_y lifetimes of all (B)Chls ranging from ∼2.5 ns (BChl b) to ∼6.5 ns (Chl a, BChl c).^116,118 Much less is known about the lifetime of the states giving rise to the Soret band. The Soret lifetimes of Chl a in a few different solvents were reported,¹¹⁹ yielding values of ∼150 fs. Comparable relaxation times were obtained after Soret excitation of Chl a (145 fs) and Chl b (160 fs) in ethanol solution.¹²⁰ 

The energy relaxation between Q_x and Q_y states was estimated to take place on a sub-100 fs timescale.^{72,119,121,122} Coherent vibrational dynamics, i.e., vibrational wavepackets in these molecules have been also investigated using the 2DES technique.^72,123,124 Before that, the nature of the excited states and involvement of vibrations have been extensively studied using high-resolution spectroscopy techniques (see Refs. 113 and 125 and references within). The lack of symmetry between the absorption and emission spectra, as well as the difference between the vibrational spectra in ground and excited states of Chls suggested the presence of vibronic mixing and that Condon and Born–Oppenheimer approximations are not valid for these molecules. More recently this was also suggested in the combined theoretical and anisotropy 2DES study of Chl a in solution.¹²² In 2020, with the help of a polarization-controlled 2DES experiment (Sec. II B 3 c), which is sensitive only to the signals from either purely electronic coherences or coherences excited via vibronically mixed transitions, vibronic mixing between the Q_x and Q_y transitions in Chl c was unambiguously identified.⁷² It was also discovered that several modes are responsible for the vibronic mixing. It seems reasonable to assume that vibronic mixing is present in all chlorophyll-type molecules as suggested by Reimers et al.,¹¹³ but this will have to be confirmed in future experimental studies. Furthermore, the effect of the vibronic mixing in chlorophylls on modulating or optimizing photosynthetic functions remains to be explored.

Phycobilins, which occur exclusively in phycobilisomes, are linear tetrapyrroles derived from the bile pigment biliverdin. Three major bilins are found in phycobilisomes: phycocyanobilin (Fig. 3) absorbing in the 600–670 nm region, phycoerythrobilin (540–570 nm), and phycourobilin (490–510 nm).¹²⁶ Isolated phycobilins in solution have photophysical properties very different from those reported for phycobilins bound in phycobilisomes.¹²⁷ The binding site in phycobiliproteins locks the phycobilins in a specific conformation, preventing the conformational disorder that exists in solution. Phycobilins in solution have low fluorescence quantum yields (<0.1). They exist in several ground state conformers which exhibit rich photophysics resulting in short excited-state lifetimes ranging from 3 to 100 ps.^127–129 In contrast, the lifetime of phycobilins bound to phycobilisomes are >1 ns, resulting in fluorescence quantum yields of ∼0.5.¹²⁶ 

Carotenoids feature a complex electronic structure, and ultrafast spectroscopy has been proven to be crucial to disentangle their excited state properties. Carotenoids (see Fig. 3 for the structure of β-carotene) are derived from linear polyenes. The characteristic structural feature is the presence of methyl groups attached to the polyene backbone. The methyls cause bending of the carotenoid backbone that is not present in polyenes.^130,131 In addition to the methyl groups, a number of carotenoids contain various terminal groups, leading to a large variability of carotenoid structures, currently counting to nearly 1000. Carotenoids are very good colorants due to their high molar absorption coefficients (ε ∼ 10⁵ M⁻¹cm⁻¹), but at the same time, they are essentially non-fluorescent. This apparent mystery was resolved in 1972 when experimental¹³² and theoretical¹³³ study of the polyene octatetraene reported the existence of a low-lying state, which was forbidden for a one-photon transition from the ground state. These two papers became a landmark of carotenoid photophysics studies, which flourished especially after the appearance of ultrafast lasers enabling sub-ps time resolution.

Following the discovery of the low-lying carotenoid dark state, the carotenoid excited-state properties were described using a three-state model consisting of the absorbing state (S₂), the low-lying dark state (S₁) and the ground state (S₀) (Fig. 4). These states are, in the C_2h symmetry point group, often used to approximate the symmetry properties of polyenes and carotenoids, denoted by A_g⁻ (S₀ and S₁) and B_u⁺ symmetry labels.¹³⁴ The same parity of the S₀ and S₁ states (both A_g⁻) has been traditionally used to rationalize the forbiddenness of the S₀–S₁ transition;^134–136 however, recent reports show that the picture is far more complicated.^137,138 This standard three-state model is often used to describe carotenoid excited states even today, though a number of experimental and theoretical studies carried out during the past two decades suggested that some additional states, which will be described below, are needed.

Within the framework of the three-state model, carotenoids absorb light via the S₀–S₂ transition characterized by a large oscillator strength in the 400–550 nm spectral region, resulting in the typical yellow-to-red color of carotenoids. The excited S₂ state is depopulated via internal conversion to the dark S₁ state, which then decays, again non-radiatively, to the ground state. The first direct measurement of the S₁ lifetime was reported by Wasielewski and Kispert,¹³⁹ yielding S₁ lifetimes of 5.2, 8.4, and 25.4 ps for the carotenoids canthaxanthin, β-carotene, and 8′-apo-β-carotenal. Direct measurements of the S₂ lifetime were not possible until the time resolution of ultrafast spectroscopy broke the 200 fs limit. Shreve et al.¹⁴⁰ obtained a 200–250 fs S₂ lifetime of β-carotene in 1991, which was at the limit of their time resolution. Later, Kandori et al.¹⁴¹ in 1994 used fluorescence upconversion with 100 fs time resolution to measure the decay of weak β-carotene S₂ fluorescence and obtained an S₂ lifetime of 200 fs. This ultrafast relaxation scheme explains the extremely low fluorescence quantum yields of carotenoids that, with a few exceptions, do not exceed 10⁻⁴.^142,143

Already the first measurement of a carotenoid S₁ lifetime¹³⁹ demonstrated that excited-state properties depend on carotenoid structure. The key structural feature determining the spectroscopic properties is the conjugation length, N. In the simplest approach, N is determined by the number of conjugated C=C bonds, but a number of experiments showed that this is valid only for linear carotenoids. If the conjugation is extended to terminal rings or even includes other groups such as C = O, a notion of effective conjugation length, N_eff, must be introduced. Thus, even though β-carotene (Fig. 3) has 11 conjugated C=C bonds, those located within terminal rings do not fully contribute to the conjugation, because the rings are twisted with respect to the plane of the linear conjugated backbone.^144–146 N_eff may vary even for carotenoids having the same conjugated system as demonstrated for renierapurpurin and isorenieratene (with the S₁ lifetimes of 6.5 and 13 ps, respectively) that differ only in positions of methyl groups at their terminal ring. Because of this, the terminal rings have different orientation with respect to the conjugated backbone, affecting N_eff.¹⁴⁷ 

Ultrafast spectroscopy has been the key tool to determine lifetimes of carotenoid excited states and their relation to the carotenoid structure. The S₁ lifetimes are readily measurable via decay of the excited state absorption from the S₁ state. The S₁–S_n transition, which for most carotenoids occurs in the 500–700 nm spectral region,¹³⁶ has an oscillator strength comparable to that of the S₀–S₂ transition, and it is a fingerprint of the S₁ state, with its shape and lifetime characteristic for each carotenoid. To this date, the S₁ lifetimes of more than 50 carotenoids occurring in nature, whose effective conjugations cover the range N_eff ∼ 8–14, have been measured.¹³⁶ It is well established that the S₁ lifetime follows the energy gap law: the S₁ energy drops with increasing conjugation, which results in shortening of the S₁ lifetime.¹⁴⁸ The longest S₁ lifetimes measured for natural carotenoids are in the 160–180 ps range for the carotenoid peridinin (N_eff ∼ 8),^149,150 but only in nonpolar solvent since peridinin belongs to the family of keto-carotenoids whose lifetimes depend on solvent polarity (see below). The carotenoid diketospirilloxanthin (N_eff ∼ 14) with the S₁ lifetime of 0.8 ps (Ref. 151) is on the other side of the S₁ lifetime range.

The measurements of natural carotenoids have been complemented by data obtained on a number of synthetic carotenoids, extending the N_eff range. At the long end, synthetic analogs of β-carotene with 15 and 19 conjugated C=C bonds (N_eff approximately 14 and 18) have S₁ lifetimes of 1.1 and 0.5 ps, respectively.^152,153 A synthetic homolog of zeaxanthin with 23 conjugated C=C bonds is the longest carotenoid whose S₁ lifetime has been measured so far. Formally, this corresponds to N_eff ∼ 22, but for such a long conjugated system, the spectroscopic difference between N and N_eff is negligible.¹⁵⁴ The N = 23 zeaxanthin has an S₁ lifetime of 200 fs.¹⁵⁵ At the short end of carotenoid conjugation, excited-state properties of β-carotene homologs with N_eff ∼ 5.5 and 3.9 were reported, yielding S₁ lifetimes of ∼300 ps and 2.7 ns, respectively;^152,153 for synthetic peridinin with N_eff ∼ 4.5 an S₁ lifetime of 2.9 ns has been reported in n-hexane.¹⁵⁶ 

The carotenoid S₁ state was also a principal target for time-resolved Raman spectroscopy. Carotenoids have strong Raman signals associated with C–C and C=C stretching modes, which in the ground electronic state have frequencies of about 1150 and 1520 cm⁻¹, respectively. The frequency of the ground state C=C stretching mode depends on N_eff and has been often used to determine N_eff.^145,146 The first time-resolved Raman measurements with picosecond time resolution, however, showed that in the S₁ state the C=C stretching mode is significantly up-shifted, yielding values as high as 1780 cm⁻¹.¹⁵⁷ This upshift, which results from strong vibronic coupling between the S₁ and S₀ states,¹⁵⁸ was later confirmed by the first time-resolved Raman spectroscopy experiments with sub-ps time resolution,¹⁵⁹ and carotenoids have become popular testing molecules in the rapidly developing field of femtosecond Raman spectroscopy (Sec. II B 3 a).^160,161 Subsequent measurements of time-resolved Raman spectra on various carotenoids confirmed the existence of the characteristic upshifted C=C stretching mode in the S₁ state, whose dynamics corresponded to the S₁ lifetimes. The upshifted C=C stretch was reported for peridinin at ∼1700 cm⁻¹,¹⁶² fucoxanthin at ∼1750 cm⁻¹,¹⁶³ β-carotene at ∼1780 cm⁻¹,^57,164–166 zeaxanthin at ∼1775–1790 cm⁻¹,^166–168 canthaxanthin and astaxanthin at ∼1770 cm⁻¹,¹⁶⁶ and spirilloxanthin at ∼1740 cm⁻¹.¹⁶⁹ Since the reported values are listed in order of increasing N_eff, it appears that, in contrast to the C=C stretching mode in the ground state, there is no direct correlation between N_eff and C=C stretching frequency in the S₁ state.

While the S₁ lifetimes exhibit a clear correlation with the effective conjugation length, the relation between carotenoid structure and S₂ lifetime is more complicated. The S₂ lifetime drops below 100 fs for some carotenoids, requiring very high time resolution in order to characterize it. The first measured S₂ lifetimes of β-carotene (200 fs)^140,141 were overestimated, as experiments with better time resolution provided sub-200 fs values.¹⁷⁰ A thorough fluorescence upconversion study of the β-carotene S₂ lifetime in a wide range of solvents¹⁷¹ showed that the S₂ lifetime also follows the energy gap law between the S₁ and S₂ states. The S₂ lifetime decreased with increasing solvent polarizability, which also decreased the S₂–S₁ energy gap. The S₂ lifetime varied from 180 fs in n-hexane to 120 fs in quinoline, because the S₂–S₁ energy gap decreases with increasing polarizability.¹⁷¹ However, this works only for the same carotenoid in solvents with varying polarizability. The dependence of the S₂ lifetime on N_eff exhibits a peculiar behavior as it decreases with increasing N_eff in the 9–13 interval, while the S₂–S₁ energy gap is predicted to increase with increasing N_eff.¹³⁴ Thus, the longest S₂ lifetime of 240 fs was reported for carotenoids with N_eff ∼ 7, while it drops to 70 fs for N_eff ∼ 13.¹⁵³ However, shortening the carotenoid even further decreased the S₂ lifetime again, yielding 160 fs for N_eff ∼ 6.¹⁵³ This anomalous N-dependence of the S₂ lifetime is one of the indications suggesting that the three-state model is not sufficient to explain all the subtleties of carotenoid excited-state dynamics, and additional states may be required to describe the relaxation dynamics and pathways (see below). However, a fully quantum mechanical model denoted as vibrational energy relaxation approach (VERA) developed recently¹⁷² explains the anomalous N-dependence of the S₂ lifetime without invoking any intermediate state(s). Instead, the relative displacement of minima of S₂ and S₁ potential energy surfaces is the key parameter determining the relaxation rate.¹⁷³ 

Vibrational cooling is the last dynamical process described within the three-state model. The S₂–S₁ internal conversion leaves molecules in a hot S₁ state, which relaxes on the sub-picosecond timescale. The S₁ vibrational cooling was first described in two papers in 2002.^174,175 These papers reported red-shifted and broadened S₁–S_n transitions at early delay times due to the population of higher vibrational states of the S₁ state. Relaxation of this band shape to the final relaxed S₁–S_n spectral profile is associated with vibrational cooling in the S₁ state that for β-carotene was observed to occur in 300–500 fs, depending slightly on solvent. Later, routine use of broadband detection combined with global fitting methods identified vibrational relaxation in essentially all carotenoids. Studies of a series of carotenoids with varying N_eff revealed a systematic acceleration of S₁ vibrational cooling with increasing N_eff, from nearly 1 ps in neurosporene (N_eff = 9) to 160 fs in spirilloxanthin (N_eff = 13).¹⁷⁶ 

Relaxation within the S₂ vibrational manifold has not been reliably identified so far. Reports targeting this process are scarce and despite earlier attempts using excitation to the higher vibrational levels of the S₂ state, this issue remains controversial. Based on fluorescence upconversion data, Akimoto et al. reported a 35–40 fs time constant,¹⁷⁷ and comparable values were obtained by Kerr-gate fluorescence spectroscopy.¹⁷⁸ Since a conical intersection has been invoked as one of the possible features of the mechanisms of S₂ depopulation,^179,180 it is likely that if the Franck–Condon region is situated close to the intersection point, a full vibrational relaxation of the S₂ state does not occur at all. Such a scenario was also suggested in a few other studies.^173,181,182 Detailed characterization of this process is nevertheless hindered by limits of time resolution, as well as by introduction of other ultrafast (sub-50 fs) relaxation channels to other dark state(s) proposed in a number of studies (see below).

Ultrafast spectroscopy was also important in determining the energy of the forbidden S₀–S₁ transition. Since the S₁ state is not directly accessible from the ground state, its energy has been for a long time a matter of considerable debate. Along with steady-state methods relying on measurements of extremely weak S₁ fluorescence,¹⁸³ or resonance Raman profiles of the S₀–S₁ transition,¹⁸⁴ a new approach based on measurement of the spectral profile of, the presumably allowed, S₁–S₂ transition within the S₁ lifetime was developed in 1999.¹⁸⁵ The S₁–S₂ transition occurs in the near-IR region spanning the 900–1800 nm range for most natural carotenoids and comparison of the S₁–S₂ and S₀–S₂ spectral profiles has been used to determine the S₁ energy of a number carotenoids.^185–187 

Even though the three-state model reasonably describes the basic dynamics of carotenoid excited states, already the first calculations using more advanced methods reported in Ref. 134 suggested that additional states are likely present in the lowest-energy manifold of excited states of carotenoids or polyenes (Fig. 4). The calculations predicted two other “dark” states occurring within the S₁–S₂ energy gap for some carotenoids with conjugation length in the 9–13 range. The first is a state with B_u⁻ symmetry, which should be located below the S₂ state for N_eff ≥ 10; the second is another state with A_g⁻ symmetry, which is expected to drop below the S₂ state only for N_eff ≥ 13.^134,188 However, these relations are valid only for the ground state geometry and more recent calculations suggest that the state ordering might change upon deviations from the ground state geometry leading to changes in bond length alternation.^189–192 

While the third A_g⁻ symmetry state likely has little effect on excited-state dynamics as it may interfere with the S₂–S₁–S₀ relaxation scheme only for very long conjugated systems, the search for the spectroscopic signatures of the B_u⁻ state has been an important part of research in carotenoid photophysics.¹⁹³ First, the B_u⁻ state should be within the S₂–S₁ energy gap for most carotenoids occurring in natural light-harvesting systems, and possible effects on energy transfer efficiency could be expected. Second, the B_u⁻ state was supposed to “cure” the anomalous N-dependence of the S₂ lifetime, which does not follow the energy gap law, though some recent studies showed that it is possible to explain it without invoking an intermediate state.¹⁷³ 

The first report claiming the assignment of a spectral feature to the B_u⁻ state appeared in 2002 and used a femtosecond spectrometer with sub-20 fs time resolution.¹⁹⁴ For the carotenoids lycopene and β-carotene, a spectral feature peaking around 1000 nm with an extremely short lifetime of ∼10 fs was assigned to an intermediate state tentatively denoted as S_x, invoking a S₂–S_x–S₁–S₀ relaxation scheme. Such an extremely short-lived intermediate state was subsequently proposed also in other carotenoids with N_eff ≥ 10.^{190,195–198} In most reports, the S_x feature was tentatively assigned to the B_u⁻ state, though an alternative hypothesis involving a state with A_g⁺ symmetry dropping below the S₂ state at the S₂ geometry has been also proposed.¹⁹⁰ A further explanation of the intermediate S_x state was proposed by Beck's group, which included changes in carotenoid configuration into the relaxation scheme.¹⁹⁹ In such a scheme, the S_x state is assigned to the energy minimum of the S₂ potential surface corresponding to a twisted configuration reached upon onset of torsional motions induced by excitation.^200,201 Despite the number of studies mentioned above, the precise mechanism of the S₂–S₁ relaxation is still a matter of debate. One of the limiting factors is the time resolution of the experiments, as the processes involving the possible intermediate state occur on sub-30 fs timescale. Also, the presence of a multitude of experimental artifacts in the pulse overlap region complicates the matter. The data from 2DES experiments addressing this issue are also controversial, and both presence^190,198 and absence²⁰² of the S_x intermediate state have been reported by 2DES for carotenoids with comparable N_eff. It should be mentioned here that coherent 2DES is highly prone to artifacts in the pulse overlap range, which are often ignored.²⁰³ 

Another intensely debated state beyond the three-state model is commonly denoted as S^*. In contrast to B_u⁻, the S^* is readily identified via its characteristic excited-state absorption, squeezed between the ground state bleaching and the main S₁–S_n excited state absorption band.¹⁹³ This spectral feature, known as the S^* signal, was first reported in 1995 in very long (N_eff. ≈ 18) derivatives of β-carotene.¹⁵² The S^* signal exhibited slower decay than the S₁ state and was assigned to the hot ground state. However, this assignment was later challenged in a report demonstrating that the S^* signal of the carotenoid spirilloxanthin bound to a light-harvesting complex serves as a precursor for triplet formation and, therefore, must be a separate excited state.²⁰⁴ These early papers identified two possible origins of the S^* signal, hot ground state and some excited state with similar properties, yet differing from the S₁ state. Although extensive research during the past 20 years has shed some light on the possible origin of the S^* state, a conclusive answer is still missing.

The S^* lifetime and its relation to the S₁ lifetime are important factors for understanding the overall carotenoid relaxation to the ground state. A number of studies provided evidence that for carotenoids with N_eff ≥ 12, the S^* lifetime is always longer than the S₁ lifetime,^{152,155,204–206} while for shorter ones the S^* signal decays with the same lifetime as the S₁ state.^181,207 This distinction provided a basis for kinetic modeling that showed that the S^* signal could be successfully modeled by a combination of contributions from a hot ground state and the S₁ state.^207–209 The model of hot ground state was further elaborated by more sophisticated modeling using the vibrational redistribution approach VERA,¹⁷² showing that for carotenoids with N_eff < 11 (such as, e.g., β-carotene) the dominating contribution to the S^* signal comes from the S₁ state, but with increasing conjugation length the hot ground state contribution prevails, because the S₁ lifetime becomes shorter than vibrational relaxation in the ground state.^210,211

While these reports seem to resolve the S^* signal origin as due to a hot ground state, it must be noted that alternative models assigning the S^* signal to a separate excited state can also reproduce the data. In these models, the S^* state is usually associated with a distinct minimum on the S₁ potential surface, corresponding to a “distorted” S₁ state.^176,205,212 Furthermore, while the hot ground state model is feasible for carotenoids in solution, it is hardly applicable for the S^* signal reported for carotenoids bound to proteins. There, S^* has been shown to serve as a precursor to triplet formation,²⁰⁴ as a minor energy transfer donor,²¹³ and recently also as a quencher of Chl a excited states,²¹⁴ or even as a state initiating photoconversion of the orange-carotenoid protein.²¹⁵ Clearly, a hot ground state does not have sufficient energy to carry out these processes. Moreover, the S^* signal in protein complexes containing carotenoids and (B)Chls may overlap with spectral features resulting from carotenoid electrochromic shifts resulting from excited (B)Chls nearby, further complicating the analysis of the S^* signal origin.^216,217

Finally, yet another state beyond the three-state model has been identified in keto-carotenoids, which contain a conjugated keto group in their structure. The keto-oxygen introduces asymmetry in the electron distribution along the conjugated chain, resulting in strong dependence of excited-state properties on solvent polarity. The polarity-dependent behavior was first reported for the carotenoid peridinin¹⁴⁹ and subsequently detected in other keto-carotenoids. In polar solvents, a new excited-state absorption band in transient absorption spectra indicates the presence of an intramolecular charge transfer (ICT) state.¹⁴⁹ Its presence in the excited-state manifold is usually associated with significant shortening of the carotenoid excited state lifetime.^149,218 Later, another spectral signature of the ICT state, stimulated emission in the near-IR region, was discovered.^150,219

The relation between the ICT and S₁ states has been a matter of considerable debate, because these two states, clearly distinguished by their specific spectral bands, decay with the same lifetime.^218,219 Therefore, a strong S₁–ICT interaction resulting in a coupled S₁/ICT state was assumed, with S₁ and ICT corresponding to two minima on the same potential surface, separated by a low barrier.²²⁰ This picture has been only recently confirmed by pump-dump-probe spectroscopy, which allowed disturbing selectively the ICT part of the S₁/ICT potential surface.^163,221

Cyanobacteria and red algae can be found in almost every terrestrial and aquatic habitat. Phycobilisomes absorb in the range of 450 to 650 nm and are the main photosynthetic antenna of cyanobacteria, red algae and glaucophytes, attached to the outer, stromal, side of the photosynthetic membrane.¹²⁶ They are composed of two similar phycobiliprotein units, α and β, to which phycobilin pigments are attached. The most common pigment–protein complexes are phycoerythrin, phycocyanin, and allophycocyanin, which aggregate into disk-shaped trimers or hexamers, forming the two major structural entities of the phycobilisome, the core and rods. The core, attached to the photosynthetic membrane, generally consists of three cylinders of stacked hexameric disks of allophycocyanin, to which a varying number of rods with a varying number of phycoerythrin and phycocyanin disks are attached.^222–224 Phycobilisomes were thought to mainly deliver excitation energy to the PSII RC,^126,223 but more recent work has shown that both PSI and PSII accept excitations from the phycobilisomes.²²⁵ This is illustrated by the phycobilisome structural model in Fig. 5, based on x-ray crystal structures of phycocyanin, allophycocyanin, and the PSI and PSII RCs of cyanobacteria.^225,226

Non-time-resolved biochemistry and spectroscopy work showed that phycobilin pigments in the rods are organized in order of decreasing excited state energy from the outer tip toward the core, and fluorescence originates almost exclusively from allophycocyanin in the core.^126,223,227 The energy gradient is created with phycoerythrin at the outer parts of the rods and phycocyanin further down the rods toward the attachment to the core. Further fine tuning of the energy gradient is achieved by pigment–protein interactions and by pigment-less linker proteins between the disks. This already suggested that efficient directional energy transfer occurs from the outer parts of the rods toward the core and then further to the RCs.

Early time-resolved fluorescence measurements on phycobilisomes from Porphyridium cruentum, having rods with both phycoerythrin and phycocyanin, seemed to verify this picture. At sufficiently low excitation pulse energy to avoid exciton annihilation, phycoerythrin fluorescence decayed with a 70 ps lifetime and the main allophycocyanin fluorescence was seen to have a rise time of 120 ps and decay of 4 ns.^106,107 Somewhat similar results were observed in Ref. 228 for phycobilisomes from the same organism. Picosecond fluorescence measurements with low-energy dye laser excitation pulses revealed ∼60 and ∼40 ps energy transfer times from phycoerythrin and phycocyanin, respectively. The shorter energy transfer time from phycocyanin to allophycocyanin in the latter study is probably a result of the different mode of detection in the two studies—as rise time in Refs. 106 and 107 and decay time in Ref. 228. The overall energy transfer from the outer parts of the phycoerythrin rods to the terminal allophycocyanin emitter was concluded to be 70 ps, in better agreement with the allophycocyanin rise time observed in Ref. 107. A transient absorption anisotropy decay of 12 ps was assigned to phycoerythrin intra-heximer energy transfer.²²⁸ Still another time-resolved fluorescence study, this time on intact cells of the red algae P. cruentum, as well as the cyanobacterium Anacystis nidulans agreed with the above results,²²⁹ and nicely illustrated the sequential energy transfer through the phycobilisome rods, to the allophycocyanin core and finally to RCs. From the temporal evolution of the fluorescence spectrum, energy transfer was concluded to occur with a non-exponential decay law with 1/e decay times of 93, 132, 156 ps for the phycoerythrin-phycocyanin-allophycocyanin-Chl transfer steps of P. cruentum. In A. nidulans, the phycocyanin-allophycocyanin-Chl steps were seen to be somewhat slower.

Isolated phycobilisomes of another red algae, Rhodella violacea, exhibited a similar energy transfer behavior as observed for P. cruentum—non-exponential energy transfer with a dominating 34 ps component from phycoerythrin and 25 ps from phycocyanin.²³⁰ 

Picosecond fluorescence and absorption studies of isolated phycobilisomes of the cyanobacteria Synechoccocus 6301 further adds to the energy transfer picture painted above. By detecting decay of phycocyanin fluorescence and rise of allophycocyanin fluorescence, it was concluded that overall energy transfer from phycocyanin to allophycocyanin occurs in 120 ps.²³¹ These results were further corroborated by picosecond absorption measurements, verifying the overall rod to core transfer time of 80–120 ps;²³² such measurements on the mutant AN 112 having rods of only one hexameric disk (instead of three) resulted in a shorter, 45–50 ps, phycocyanin decay time, further supporting the interpretation of rod-core energy transfer time. Energy transfer from the main pool to the final allophycocyanin emitter, transferring energy to the RC, was concluded to occur in 50 ps. Intra-disk energy transfer, both in the rods and the core, was concluded to be very fast, <10–15 ps.^232,233 Similar results, <8 ps intradisk and 25 ps rod interdisk transfer, were reported by Glazer and co-workers in Ref. 234.

Much more recent measurements have been performed with higher time resolution and more sophisticated analysis. Thus, with the help of femtosecond transient absorption and picosecond fluorescence streak camera measurements and global analysis using a compartment model of the phycobilisome pigments, energy transfer was studied in isolated phycobilisomes of Synechocystis PCC 6803.²³⁵ It was found that intra-hexamer energy transfer in both phycocyanin and allophycocyanin occur within ∼2–5 ps, intra-rod equilibration 15 ps, overall rod to core transfer 50–70 ps, equilibration within an allophycocyanin cylinder 13–19 ps, and the slowest process is intercylinder equilibration in the core, 75–200 ps.

A recent whole cell time-resolved fluorescence study of Synechococcus WH 7803²³⁵ provides a detailed view of the overall energy transfer in the phycobilisome and trapping by the RCs. For Synechococcus, WH 7803 interdisk energy transfer between two different spectral forms of phycoerythrin was found to occur in 36 ps, from phycoerythrin to phycocyanin energy flows with a major fast time constant of 15 ps and a minor one of 90 ps. Energy transfer from phycocyanin to the allophycocyanin core is characterized by a 32 ps time constant and equilibration among the two allophycocyanin spectral forms occurs with a 25 ps time constant. The fastest step is a branched transfer from the low-energy allophycocyanin form (APC 680) to PSI and PSII with a combined time constant of 7 ps (11 ps to PSI and 20 ps to PSII). This study can be compared with that on P. cruentum performed 34 years earlier by Yamazaki and co-workers.²²⁹ The phycobilisomes of the two organisms are similar in that they both contain the three pigments phycoerythrin, phycocyanin, and allophycocyanin and we can see that the overall rod-to-core energy equilibration time is also similar, 100-ps timescale. However, the last energy transfer step, from the allophycocyanin core to chlorophyll pigments in the membrane, appears to be much faster in Synechococcus WH 7803, 7 ps vs 156 ps in Porphyridium.

What ultrafast spectroscopy taught us about cyanobacteria and red algae

Taken together, the measurements performed over 40 years, on both isolated phycobilisomes and intact cells of several different species firmly illustrate the function of phycobilisomes—light harvesting in the visible part of the spectrum, picosecond timescale directional energy transfer from high-energy pigments located at the peripheral edges of the phycobilisome toward the core attaching the phycobilisome to the membrane, and finally efficient energy transfer to chlorophyll pigments in the thylakoid membrane. The sequence is completed with an overall efficiency of >95%.²³⁶ It is also interesting to notice that early measurements at the outset of time-resolved studies of phycobilisome energy transfer could present a correct picture of the overall dynamics, and that more recent measurements with higher time resolution and more sophisticated analysis schemes have essentially confirmed the measurements performed 35–40 years ago, but also yielded a more detailed picture of the internal energy transfer dynamics of the phycobilisome. From the results discussed here it appears that the phycobilisome is a very efficient light-harvesting antenna, as has also been discussed in much more detail elsewhere.^126,236 One might ask the question why evolution disposed of phycobilisomes in green algae and plants? In Ref. 236, it is speculated that the reasons might not have been related to efficiency of light harvesting, but to photosynthetic membrane structure and the separation of PSI and PSII to different regions in the thylakoid membrane of these more advanced organisms.

Green bacteria are found mostly in hot springs, in sulfur-rich environments, often living at low light intensities. Green sulfur bacteria have been found in depths of up to 145 m in the Black Sea, and a species has been found living near a black smoker off the coast of Mexico at a depth of 2500 m. At this depth, the bacterium lives of the dim glow from the thermal vent since no sunlight can penetrate there. Chlorosomes (Fig. 6) are the main light-harvesting antenna of green sulfur bacteria (Chlorobi), green filamentous bacteria (Chloroflexi), and phototropic acidobacteria (Candidatus Chloracidobacterium). They contain BChl c, d, and e as the main light-harvesting pigments, which aggregates into large supramolecular complexes that may organize into different shapes, e.g., lamellae, tubes, spirals, determined by details in the substitution pattern of the BChl molecules. Unlike all other photosynthetic complexes, the organization and functional tuning of the pigments are controlled by long range pigment–pigment interactions, rather than pigment–protein interactions. Thus, within the BChl aggregates there are no or very little protein, but there are other molecules, carotenoids, and quinones with light-harvesting, redox control and protective functions; these molecules may also assist in the aggregation of the BChls. The BChl aggregates along with carotenoids and quinones are enclosed in an ellipsoidal sac of a lipid monolayer membrane, 100–200 nm long and 10–60 nm wide and thick. This membrane holds a range of proteins, often with poorly known (or even unknown) function. The most important protein is the CsmA protein, binding BChl a molecules and forming the baseplate on one side of the chlorosome. As we will discuss in more detail below, the baseplate is the conduit of energy flow from the internal chlorosome BChls to the next energy acceptor. In Chlorobi and Candidatus Chloracidobacterium thermophilum this is the Fenna–Matthews–Olson (FMO) pigment–protein complex, whereas in Chloroflexi it is the membrane bound B808–866 complex. FMO and B808–866 then finally transfer the energy to the RC pigments. See, for example, the following references for more information about chlorosome chemical composition and structure:^237–244 

The FMO complex is the first chlorine-containing protein complex to have its three-dimensional structure resolved to high resolution and named after the scientists who first identified the complex, Olson,^245,246 and performed the structural work, Fenna and Matthews.^247,248 The protein is a homo-trimer holding seven BChl a molecules per monomer; a recently identified eighth BChl a is located at the interface between the monomers.²⁴⁹ Due to the relatively small number of pigment molecules, the FMO protein has become one of the best studied photosynthetic proteins and somewhat of a test ground for experimental and theoretical energy transfer studies.^97,250

Green filamentous bacteria (Chloroflexi) lack the FMO protein that in green sulfur bacteria interface the chlorosome to the photosynthetic membrane and RC complex. Instead, there is an integral membrane protein resembling a hybrid between the LH1 and LH2 protein complexes of purple bacteria. Electron microscopy studies of the green filamentous bacterium Roseiflexus castenholzii, which lacks chlorosomes, showed that it consists of ∼15 α/β subunits surrounding the RC and holding three BChl a and two carotenoid molecules per subunit.^251,252 In R. castenholzii the BChl a molecules give rise to two absorption bands, one at 800 nm (B800-like of LH2) and another at 880 nm (LH1-like); in Chloroflexus aurantiacus, this protein complex has a bit different absorption spectrum, B808–866.²⁴² 

The aggregation of BChl molecules in chlorosomes and the variation of aggregate size and shape, discussed above, could suggest ultrafast intra-chlorosome energy transfer (exciton relaxation) as well as considerable variation in the energy transfer dynamics depending on bacterial species of origin of the chlorosome. The large number of studies of chlorosome energy transfer by many groups seem to verify this expectation. Rather than discussing all individual studies one by one we will identify common results and general trends, and where there are significant differences we try to understand the reasons. Experimental methods of choice to achieve the expectedly necessary femtosecond temporal resolution is in most studies transient absorption or two-dimensional electronic spectroscopy. All studies agree that the energy relaxation occurs over a wide time range. The fastest processes on the few-hundred fs timescale are assigned to relaxation from high energy exciton states to lower-lying states.^{243,253–255} In addition, all studies agree that there are slower, 10–20 ps, processes, but other, both faster (∼1 ps) and slower (∼30 ps), dynamics are reported. These processes are generally assigned to relaxation among low-lying exciton states,^77,253,256 or alternatively expressed, between exciton levels located on different BChl layers of the chlorosome lamellae. The spread in lifetimes for the slow components is perhaps not surprising considering that studies were performed on chlorosomes from both Chlorobi and Chloroflexi containing both BChl c and BChl e, with the possible variation in BChl c/e aggregate structure. A 2D electronic spectroscopy²⁵⁴ work showed that incoherent exciton diffusion within and between coherent domains occurs on a sub-100 fs timescale. Most of the distribution of exciton energies are sampled within 200 fs, but minor evolution of the 2DES spectrum on a slower timescale shows that there are other slower energy transfer processes, as suggested in earlier studies.

Theoretical simulations of exciton dynamics point in the same direction as the experiments regarding the ultrafast part of the relaxation processes. Thus, exciton delocalization was concluded to be complete in 100–200 fs (Refs. 255 and 257) and energy relaxation in the exciton manifold in ∼50 fs. For excitonically coupled pigment molecules, the exciton delocalization size is a quantity reporting on properties of the aggregate (e.g., pigment–pigment coupling, energy disorder, dynamic disorder, etc.).^258–260 The photoinduced bleaching in the BChl c Q_y absorption band of chlorosomes from C. aurantiacus was observed to be 7–8 times greater than that of the BChl a band of the chlorosome. This was interpreted as proof of exciton delocalization over many BChl c molecules. These results were further analyzed with exciton theory using an aggregation model of the BChl c molecules²⁶¹ and found to correspond to an aggregate size of 24 BChl c molecules arranged as a tubular aggregate of six linear chains of four BChl molecules with an exciton delocalization size (defined as inverse participation ratio of the density matrix) of 7.4 at room temperature.^260,262

The BChl a baseplate of the chlorosome is the gateway out for the energy from the BChl c (d or e) interior of the chlorosome. The transfer can be timed by measuring the appearance of BChl a excitations, or the overall decay of BChl c excited states. Since the baseplate is an integral part of the chlorosome, this time should be obtainable from measurements on isolated chlorosomes, but it is interesting to compare it with the corresponding measurements on membranes or intact cells to see if the interactions with the photosynthetic membrane and proteins attached to it (e.g., FMO) somehow influence the energy transfer. BChl c(d, e) to baseplate BChl a energy transfer has been studied in both Chlorobi and Chloroflexi isolated chlorosomes. Despite a significant spread of measured lifetimes in different studies, it seems that the BChl c(d, e) to baseplate energy transfer is faster in Chloroflexi, 8–40 ps (Refs. 253, 263, and 264) than in Chlorobi, ∼70–140 ps.^{77,253,256,263} In some studies, for chlorosomes of both taxa, a minor 2–10 ps component in the BChl c(d, e) to baseplate energy transfer was also detected.^77,256,264 The faster energy transfer in Chloroflexi could perhaps be related to the bigger chlorosomes of the Chlorobi taxa.²⁴⁴ Alternatively, BChl that is not BChl c in the chlorosome could also lead to a different spectral overlap and slower energy transfer. Three different studies on whole cells of C. aurantiacus all report 15–16 ps for the BChl c to baseplate energy transfer.^263,265,266 Thus, there appears not to be a significant difference in the transfer time measured in isolated chlorosomes and whole cells.

There are relatively fewer direct studies of intra-baseplate energy transfer studies, but one report²⁵⁶ assigns a transient absorption anisotropy decay at 807 nm in chlorosomes of Chlorobaculum tepidum of ∼1 ps to energy transfer between BChl a molecules in the baseplate. A 2DES study on chlorosomes of the same bacterium at 77 K could resolve relaxation processes through four exciton states, originating from four BChl a molecules that likely interact strongly, on the timescale of <1 ps to 20 ps. The faster ∼1 ps relaxation within the higher-lying states, and the slow ∼20 ps among the two lowest states.²⁶⁷ Thus, it appears that energy relaxation and equilibration within the BChl a baseplate is fast, particularly at room temperature. This, and the fact that BChl c(d,e) to BChl a transfer is much slower in all chlorosomes makes it difficult to observe this process except with direct BChl a excitation.²⁶⁷ It is worth noting that the electronic structure found in the baseplate is similar to that found in the FMO complex and the RC core antenna. This led authors to propose the feasibility of “lateral” energy transfer through the system, where excitation does not necessarily relax to the lowest state in the baseplate or FMO, before it is transferred to FMO or RC, respectively.²⁶⁷ 

As mentioned above, chlorosomes contain carotenoids having light-harvesting and photoprotection functions.^11,268 A few studies have revealed ultrafast, ∼100 fs energy transfer from the carotenoid S₂ state to BChl c with 50%–80% efficiency.^268,269 Carotenoid to BChl c energy transfer was also studied in synthetic BChl c aggregates^270,271 and similar to the work of Psencik et al.²⁶⁹ it was found to occur from the S₂ state, but somewhat slower, ∼500 fs, and less efficiently 30%–40%. The carotenoid molecules were also found to influence the aggregation of the BChl c molecules in the synthetic aggregates.²⁷¹ 

The group of Struve (and later together with Savikhin) were some of the first to study FMO energy transfer dynamics. In two early reports, using few-ps dye laser pulses in one-color transient absorption measurements,^272,273 anisotropy decays of 2.3–4.7 ps were measured and interpreted as inter-monomer energy transfer in the trimeric FMO protein. The reason for this interpretation was that intra-monomer energy transfer (or exciton relaxation) between the strongly interacting BChl c molecules was considered too fast to be resolved with the picosecond resolution of the experiment. With the advent of much shorter Ti:Sapphire laser pulses, these experiments were repeated with better temporal resolution,^274,275 and faster relaxation processes could be resolved. Thus, 100–900 fs lifetime components in isotropic and anisotropic transient absorption measurements were interpreted as energy equilibration between higher lying exciton states, and a slower anisotropy decay phase of 1.4–2.0 ps was tentatively assigned to equilibration among the lowest energy pigments.^274,275 Further broadband measurements of transient absorption spectra with several different excitation wavelengths within the FMO absorption band at 17 K resulted in a level to level relaxation model.^96,276 Decay from upper to intermediate exciton states was found to occur in ∼30–100 fs, while the further decay to the lowest exciton states was concluded to be considerably slower, ∼0.6–2.5 ps. At room temperature, these processes, in particular the decay to the lowest states, would presumably be faster. There is a good agreement in reported time constants from all studies, but the early assignment of the few-ps processes to inter-monomer energy transfer^272,273 now appears to be intra-monomer decay between the lowest exciton levels. However, we need to keep in mind that the early measurements were performed at room temperature, whereas the later ones were at 17 K, so the agreement may still be a coincidence. As a concluding remark, in Ref. 96 it was mentioned that “a major impediment to molecular level interpretations of such experiments is the absence of a reliable theory for protein site effects on pigment properties…” and “New strategies for determining the antenna electronic structure (e.g., by independent nonlinear optical techniques) would be valuable.”²⁷⁶ 

Nineteen years, later this was achieved with the help of 2D electronic spectroscopy (2DES). Thus, with 2DES and full polarization control at 77 K an accurate description of electronic structure, with state to state correlation, and energy transfer dynamics was obtained.⁹⁷ The recently discovered eighth BChl a molecule could in addition be spectroscopically identified. The energy transfer scheme obtained from the 2DES measurements is similar to that generated previously by more traditional transient absorption measurements,^276,277 but the level-to-level scheme is more detailed. Thus, the energy transfer dynamics corresponds to a picture of energy flow between the eight BChl c molecules of the FMO monomer.

The energy transfer dynamics discussed above were obtained for FMO proteins isolated from two different green sulfur bacteria, C. tepidum and Prosthecochloris aestuarii, so it cannot be excluded that some differences in the reported dynamics originates from this difference in origin.

The ideas that superpositions between the excitons may play a role in photosynthetic energy transfer dates back to the 1960s.²⁷⁸ The observation of coherent oscillations in transient absorption anisotropy measurements of the FMO complex of C. tepidum at 19 K was a first hint that such effects may be present.²⁷⁹ The oscillations had a period of 220 fs, and a dephasing time of 140–180 fs and were almost absent in the magic angle signal; they were of much higher amplitude when the excitation pulse spectrum covered both the 815 and 825 nm bands of the absorption spectrum, compared to when only one of the bands was excited. These features led the authors to conclude that the oscillations were a result of quantum beats between two exciton states, rather than due to vibrational coherence. However, the authors did not discuss possible implications of these findings for energy transfer in the FMO protein. Later, in 2007, oscillations were claimed to persist for up to a picosecond in 2DES measurements on the FMO protein at 77 K,⁸³ and interpreted as electronic coherences. These results were taken as the signature of wavelike characteristics of energy transfer, explaining its high efficiency. This has been shown to be incorrect in later studies. The topic is discussed in somewhat more detail in Sec. III H.

Energy transfer within the B808–866 protein in chlorosome-membrane complexes from C. aurantiacus were studied in two-color pump-probe experiments at room temperature.²⁸⁰ The B808–B866 energy transfer time was found to be 2 ps, somewhat slower than B800–850 transfer in LH2 (0.7 ps at room temperature, see Sec. III D 2 on purple bacteria and Ref. 281). A somewhat longer transfer time, 5 ps, was obtained from time-resolved fluorescence measurements.²⁸² Transient absorption measurements on the closely related B800–880 complex of R. castenholzii²⁸³ produced the same order-of-magnitude lifetime of 2.3 ps. For B808–866, no intra-B808 energy transfer was found, in contrast to LH2, where intra-B00 transfer was found to proceed on the ∼0.5 ps timescale.²⁸¹ Intra-B866 dynamics was found to be characterized by a time constant of 350–550 fs, again slower than in LH2 and LH1 of purple bacteria.²⁸¹ Carotenoid to BChl a energy transfer was studied in the B800–880 complex of R. castenholzii; for the two dominating carotenoids (methoxy/hydroxyl-γ-carotene and methoxy-keto-myxocoxanthin) S₂ to BChl transfer was found to occur with 38 (17)% efficiency and transfer times of 180 (430) fs, whereas S₁ to BChl transfer proceeds with an efficiency of 19 (17)% and transfer time 10.4 (8.3) ps. Both carotenoid-to-BChl a energy transfer channels in this complex are thus less efficient and slower than in LH2 complexes (see below).

From the accounts above, we now have some insight into the energy transfer dynamics within individual light-harvesting complexes of green bacteria, but we would also like to know how these complexes are coupled together and the resulting dynamics of energy flow between them and into the RC. For this, studies of energy transfer of intact organisms or at least photosynthetic membranes containing an intact functioning PSU are required. This implies measurement of decay and rise of excited state populations of all the LH complexes (and RC) of the PSU. An efficient energy transfer chain necessarily implies spectral overlap, i.e., spectral congestion and, from an experimental point of view, difficulties to achieve selective excitation and probing. In a PSU, there is in addition often spectral overlap between antenna and RC pigments, further adding to the complexity. As already mentioned above, energy transfer to the RC is often rate limiting, making it difficult to access the fastest CT processes in RCs from measurements on intact organisms. We will discuss this point further in the sections on RC processes.

The difficulties to resolve the energy transfer steps between LH complexes appear to have been rather difficult in intact green sulfur bacteria. Apart from a very recent 2DES study on C. tepidum,²⁸⁴ no detailed reports are available. In this study, in addition to intra-complex energy transfer, chlorosome to FMO energy transfer was reported to occur with a time constant of 70 ps, and FMO to membrane pigment transfer with a dominating lifetime of 17 ps. The efficiency of the FMO to membrane energy transfer step was reported to be ∼75%,²⁸⁴ considerably higher than the 40%–50% reported from studies of a RC-FMO complex or membrane preparations (see, e.g., Refs. 242, 285, and 286). In Ref. 285, 66% of the FMOs of the RC-FMO preparation were inactive in energy transfer, and the 34% of the FMOs that did transfer energy to the RC did so with 76% efficiency. Observed differences could suggest that the isolation procedures of the complex and membrane preparations adversely affected the coupling between FMO and the photosynthetic membrane. Additionally, measurements at physiological temperatures through the intact PSU are required to provide the final answer regarding the energy transfer efficiency.

The direct attachment of chlorosomes to the photosynthetic membrane in green filamentous bacteria, without an interfacing FMO complex, makes energy flow through the antenna system less complex. Energy leaves the chlorosome through the BChl a baseplate and enters the membrane through the RC core complex (B808–866 in C. aurantiacus), with direct access to the RC.^251,252 This means that knowing the ultrafast internal energy dynamics of chlorosomes and the RC core complex, measurements of decay of baseplate excitations, or rise time of core complex excited states reflect the overall energy transfer to RC-coupled pigments. This was done already 30 years ago in three different time-resolved fluorescence measurements,^263,265,266 all showing that the overall energy transfer time is ∼40 ps. The further trapping of the energy by the RC in C. aurantiacus was reported to be 43,²⁶³ or 70–90 ps.²⁶⁵ In Ref. 266, the RC trapping could equally well be characterized by a single exponential decay of B866 fluorescence with 140 ps lifetime, or a double exponential with the lifetimes 70 and 200 ps of approximately the same amplitude. These two lifetimes agree perfectly with the lifetimes reported in Ref. 265 for open (70–90 ps) and closed (secondary electron acceptor Q_A reduced) RCs (180–200 ps), suggesting that the sample in Ref. 266 had a mixture of open and closed RCs. These three studies together show that the overall energy transfer from chlorosomes to the RC core antenna in green filamentous bacteria proceeds with a time constant of ∼50 ps and further trapping by the photochemically active RC takes about the same time. It is important to realize that the trapping time does not represent direct energy transfer from B866 to the primary electron donor in the RC, P, but the effective time it takes for the RC to secure the energy through electron transfer to the primary pheophytin electron acceptor. From modeling of the measured kinetics, it was concluded that the excitation from the antenna visits the special pair (P) in the RC five to ten times before primary charge separation occurs, because back energy transfer from P to B866 is about ten times faster than forward energy transfer.²⁶⁵ This type of energy dynamics is called trap-limited, in contrast to diffusion-limited where the overall energy diffusion to the RC is the rate limiting process. We will see that different types of organisms represent either of these two cases, depending on the organization of the antenna and antenna-RC interactions.

What ultrafast spectroscopy taught us about photosynthetic green bacteria

We have discussed two types of green bacteria—Chloroflexi and Chlorobi. The PSU of Chlorobi consists of a light-harvesting apparatus composed of chlorosomes with an integral baseplate connecting it to the FMO complex, which channels the energy collected by the more peripheral parts of the antenna into the RC-core antenna complex. In Chloroflexi bacteria, a membrane-bound B800–866 (or B800–880), reminiscent of purple bacterial LH2, replaces the FMO complex. Apart from that, the organization of the Chlorobi and Chloroflexi PSUs is similar. The overwhelming number of BChl pigment molecules in chlorosomes, as compared to the other antenna proteins and RC, combined with partially overlapping absorption spectra has made studies of energy transfer dynamics in intact green bacteria particularly challenging. Thus, most of the published results are for individual pigment–protein complexes. Nevertheless, the following picture emerges from the available results:

• Densely packed and strongly interacting BChls in chlorosomes result in excitons delocalized over several molecules (the exact number is still debated), similar to what has been found for LH2 and LH1 of purple bacteria (see below).

• Within a chlorosome, relaxation from high-energy exciton states to lower-lying states within a BChl layer of a lamella occurs on the few-hundred fs timescale, while relaxation between low-lying exciton states on different BChl layers is slower, 1–10 ps.

• Energy absorbed by the BChls within the chlorosome ends up in the baseplate of the chlorosome within ∼15–100 ps, the longer times for the bigger chlorosomes of Chlorobi bacteria. Energy relaxation and equilibration within the baseplate proceeds on the 20 ps timescale.

• Thanks to the relatively few (eight) BChl a molecules of the FMO complex in Chlorobi bacteria an accurate description of electronic structure and energy transfer dynamics with site-state correlation has been obtained. Energy equilibration between higher lying exciton states proceeds on the ∼50 fs timescale, and relaxation to and equilibration among low-lying exciton states takes 1–2 ps. The more than 10-year long debate on long-lived electronic coherence in FMO and its role for energy transfer efficiency appears to have been settled with the conclusion that the coherences are of vibronic origin, and thus no different than those observed in many other pigment systems with ultrashort pulse excitation (for details, see Sec. III H on coherences in photosynthetic systems, below).

• In Chloroflexi bacteria, lacking FMO, chlorosomes are interfaced to the RC-core antenna complex in the membrane by the B800–866 (or B800–880) complex. Energy transfer within this complex is similar to that in LH2 of purple bacteria—∼2 ps energy transfer from B800 to B866/B880 and few-100 fs dynamics within B866/B880.

• The difficulties in resolving the energy flow through the complete PSU of a green bacterium were mitigated with the help of 2DES experiments on intact C. tepidum. In addition to the intra-complex processes, the missing information on inter-complex energy transfer was achieved—chlorosome to FMO 70 ps; FMO to membrane pigments 17 ps at 77 K.

• From all this, the overall energy transfer time from the peripheral antenna (chlorosomes) to RC core antenna in the membrane appears similar to the corresponding overall transfer time in purple bacteria (∼60 ps), discussed below, but the overall efficiency has been reported to be lower. The reason may partially be due to the very different organization of the PSU. Alternatively, more accurate experiments at physiological conditions might end up showing more efficient energy transfer.

Photosynthetic purple non-sulfur bacteria are distributed widely in natural habitats, particularly in those with large amounts of soluble organic matter, such as swamps, wastewater ponds, coastal lagoons, and waste lagoons. Purple sulfur bacteria, on the other hand, are generally found in illuminated anoxic zones of lakes and other aquatic habitats where hydrogen sulfide accumulates and also in “sulfur springs” where geochemically or biologically produced hydrogen sulfide can trigger the formation of blooms of purple sulfur bacteria. Anoxic conditions are required for photosynthesis; these bacteria cannot thrive in oxygenated environments.

The photosynthetic unit of purple non-sulfur bacteria (in the following, just purple bacteria) consists of two different light-harvesting complexes, a peripheral LH2 antenna and a core LH1 antenna coupled to the RC. The high-resolution structures of two different LH2 complexes have been determined—LH2 of Rhodoblastus acidophilus (formerly Rhodopseudomonas acidophila) was solved in 1995 (Ref. 287) (Fig. 7) and that of Phaeospirillum molischianum (formerly Rhodospirillum molischianum) a year later.²⁸⁸ Both are a circular arrangement of transmembrane pairs of single helix α/β polypeptides, nine for R. acidophilus and eight for P. molischianum. Each α/β dimer holds three BChl a molecules and one carotenoid, thus in total 27 BChls and nine carotenoids for R. acidophilus (24 and 8, respectively, for P. molischianum). For R. acidophilus (P. molischianum) nine (eight) BChl a molecules are organized in a ring, at a Mg–Mg distance of 21 Å (very similar for P. molischianum), with the BChl macrocycles almost parallel to the photosynthetic membrane. This ring of nine (eight) BChl molecules is associated with the 800 nm absorption band and therefore termed B800. The relatively long distance between the BChl molecules in the B800 ring leads to weak BChl–BChl interaction, estimated to ∼25 cm⁻¹ (see, e.g., Refs. 281 and 289 and references therein). Eighteen (R. acidophilus), or 16 BChls (P. molischianum) collectively absorbing at ∼850 nm are organized in another circle (B850), with molecular planes perpendicular to the membrane. The Mg–Mg distances are much shorter in B850, ∼9 Å, leading to much stronger BChl–BChl interaction estimated to be on the order of 300 cm⁻¹ (see, e.g., Ref. 281 and references therein). The carotenoid molecules run between the α and β polypeptides, perpendicular to the plane of the membrane.

LH1 has a similar arrangement of α/β polypeptide transmembrane helices as LH2, with the RC encircled. Structures of LH1-RC complexes of several purple bacteria species have been determined, exhibiting a variation in size and shape (Fig. 8)—circular²⁹⁰ and elliptical²⁹¹ encapsulating one RC, and S-shaped embracing two RCs.²⁹² The elliptical LH1-RC complex of Rhodopseudomonas palustris²⁹¹ has an opening in the sequence of 15 α/β polypeptide pairs, and the S-shaped LH1-RC complex of Rhodobacter sphaeroides²⁹² has similar gaps at the ends of the “S.” These gaps in the LH1 continuity have been shown to be associated with a pigment-free protein, helix W,²⁹¹ or pufX²⁹² and to be involved in the trafficking of quinones in and out from the RC. The LH1-RC complexes have one ring of BChl a molecules, perpendicular to the membrane plane, i.e., similar to B850 of LH2. The pairwise BChl Mg–Mg distances are similar to those in B850 leading to an estimated interaction of ∼500 cm⁻¹ (see, e.g., Ref. 281 and references therein). LH1 of several purple bacteria has a main absorption band at 875 nm (B875) related to the BChl a molecules. The LH1-RC complexes bind one carotenoid molecule per α/β dimer, similar to LH2. Structures of the three known types of LH1-RC complexes are shown in Fig. 8.

During a period of approximately fifteen years from the mid-1980s to the end of the 1990s, with a peak during the years following the publishing of the LH2 structure,^287,288 intense research by several groups contributed to a comprehensive and detailed picture of the energy flow through the antenna network of purple bacteria. Already before the high-resolution LH2 structures became available, a significant amount of data on excitation dynamics had been accumulated. When these data were combined with the structural information a dynamics-structure-function picture of the light-harvesting processes could rapidly be developed and further refined in the following years.²⁸¹ Much more recent work and application of new experimental methods have added further important pieces of information to the description. Below, we summarize how the picture of purple bacterial light harvesting developed.

We start by considering the energy transfer process in LH2 that contributes to the downhill energy flow toward the RC, B800–B850. This is one of the most studied and best characterized processes in photosynthetic light harvesting. Already, early picosecond absorption²⁹³ and fluorescence²⁹⁴ measurements on Rb. sphaeroides chromatophores at room temperature indicated that this process occurs with a time constant of ∼1 ps. Several later measurements with shorter sub-ps and fs pulses provided a more precise value of the time constant of this transfer step—0.7 ps at room temperature.^295–302 These measurements were in addition performed on LH2 complexes of different purple bacteria species, showing that this energy transfer step is species independent (within experimental error of the measurements). In order to provide insight into the mechanism of B800–B850 energy transfer, a series of LH2 mutants with a successively blue-shifted B800 absorption band were studied. It was found that the B800–B850 energy transfer time gradually increases with increasing energy separation between the B800 and B850 bands, in qualitative (but not quantitative) agreement with Förster spectral overlap calculations.^303,304 The temperature dependence of B800–B850 energy transfer was also examined and the transfer time was found to increase to 1.2 ps at 77 K (Refs. 300, 301, and 305) and 1.5 ps at 4 K.³⁰⁰ This is the expected trend for Förster energy transfer, due to a decreased spectral overlap at lower temperature. However, the measured transfer times were found to be considerably shorter than the calculated Förster transfer times, based on a quantitative spectral overlap calculation, and the temperature dependence was not reproduced by the calculations.³⁰⁰ To resolve this issue, it was proposed that spectral overlap between B800 and the upper exciton component of B850 provides an additional pathway of energy transfer.^300,306,307

From the LH2 structure,^287,288 distance between the BChl molecules of the B800 ring, and an estimate of nearest neighbor interaction between B800 molecules (∼25 cm⁻¹),²⁸¹ it can be concluded that B800 excited states are localized, but the interaction is still sufficiently strong for fast energy transfer. The first hint that it actually occurs comes from the observation that B800 fluorescence at 4 K is depolarized.⁹⁵ In a time-resolved measurement, energy transfer within the B800 ring can be detected in two different ways, as a decay of time-resolved anisotropy as the excitation moves around the ring, or as an excitation wavelength dependent change of isotropic absorption or fluorescence intensity, if the absorption band is inhomogeneously broadened. In the latter case, excitation in the blue wing of the absorption spectrum would lead to a decay at short wavelengths and a corresponding rise of signal at long wavelengths. Both these features have been observed for B800. The first time-resolved absorption anisotropy study of B800 of two different LH2s of Rb. sphaeroides and R. palustris at room temperature revealed an ∼1 ps anisotropy decay,²⁹⁶ a similar anisotropy decay time was found for B800 of R. acidophilus.³⁰¹ At 77 K, a shorter, ∼300 fs, depolarization time was observed,^301,308 and master equation simulations describing the energy transfer as incoherent hopping in the ring of B800s, showed that the measured B800 dynamics could be described as energy transfer with a characteristic average nearest-neighbor pairwise transfer time of 0.35 ps. Several isotropic transient absorption measurements at 77 K revealed an ∼500 fs relaxation time interpreted as energy transfer from blue to red B800s.^{301,305,306,308,309} Considering all measured dynamics of the B800 band, it was concluded that “energy transfer within the B800 ring can be understood on the basis of the Förster equation for energy transfer in the weak coupling limit,” and “a neighbor-to-neighbor transfer time of 0.5–1 ps would be more than sufficient to account for the experimental observations.”²⁸¹ This includes the longer transfer time at room temperature,²⁹⁶ as well as a factor of two longer B800-B800 transfer time in LH2 of P. molischianum.³¹⁰ 

Purple sulfur bacteria also contain LH1 and LH2 complexes, but much less is known about their structure, spectroscopy, and light-harvesting processes. Allochromatium vinosum (formerly Chromatium vinosum) is an interesting, somewhat peculiar species. Instead of the single B800 absorption band of purple non-sulfur bacteria, discussed so far, LH2 of the purple sulfur bacterium Alc. vinosum exhibits a split B800 absorption band.^311,312 Different explanations to this have been offered; one assumes that there are two spectrally different LH2 complexes, impossible to separate with biochemical methods. Another explanation suggests that there are two different B800 BChl sites within the same LH2 complex, due to different pigment binding pockets. A third hypothesis is that the B800 BChls are organized in excitonically coupled dimers, as a consequence of reduced inter-chromophore distance, which results in a splitting of the absorption band.³¹² Non-conclusive and mutually conflicting results for these hypotheses have been presented.^312,313 With the help of 2DES, this unsatisfactory state of affairs was resolved.³¹⁴ The detection of a cross-peak between the two diagonal B800 peaks conclusively showed that there is excitonic coupling between the BChl molecules of the two B800 sub-bands and proved that the hypothesis of two spectrally different LH2 complexes does not hold. Nevertheless, it was concluded that this finding does not exclude that the BChl molecules of the two B800 bands could have different site energies.³¹⁴ 

From the short BChl–BChl distances, ∼9 Å, in the B850 ring, it is immediately clear that the interaction between BChls is strong—with some variation of the precise value, it has been estimated to be on the order of 300 cm⁻¹, and the inhomogeneous broadening of the same order of magnitude.²⁸¹ This doubtlessly leads to excitonic delocalization of the excited states. The degree of exciton delocalization was a controversial issue during a few years following the publication of the LH2 structure.

There are several methods that can be used to access the size of the exciton. The magnitude of photobleaching (ground state to one-exciton transition), or induced absorption (one-exciton to two-exciton transition) of the antenna complex, as compared to the monomer pigment, is an apparently straightforward measure that has been used to conclude that the exciton is delocalized over a large part of the B850 and LH1 rings, or most likely the whole ring.^315,316 The same photobleaching method, but with the B820 BChl dimer as reference was used to avoid sources of experimental uncertainties, and an exciton coherence length of 5 BChls was obtained.³¹⁷ The transient absorption spectrum and the energy difference between the peaks of ground state bleaching and induced absorption is another measure³¹⁸ that was used to estimate the size of the B850 exciton to 4 ± 2 BChls at 2 ps after excitation.³¹⁹ This was confirmed by more sophisticated exciton calculations including diagonal energy disorder and fitting to experimental transient absorption spectra of B850 in the temperature range 4–296 K. At 1.5 ps after excitation, a relaxed exciton delocalization length of 4 ± 1 BChl molecules, independent of temperature, was obtained.³²⁰ Using multilevel Redfield theory including both static and dynamic disorder (exciton–vibrational coupling), exciton dynamics of B850 was investigated.²⁵⁸ The exciton size was found to be time dependent, delocalized over a large part of the B850 ring at very early times and have a relaxed size of ∼4 BChl molecules, again confirming this value as the relaxed exciton size of the B850 exciton. Super-radiance, or radiative rate, is a third method that has been used to access the exciton delocalization of B850 and LH1, and a very similar value, a few BChls, as in the analysis of transient absorption spectra was obtained.²⁸¹ Together, these results show that the relaxed exciton size in B850 and LH1 is ∼4 BChl a molecules, largely independent of temperature. The exciton delocalization is time dependent and at very early times depending on excitation conditions it may span most of the B850 and LH1 rings.²⁵⁸ 

The exciton within the B850 and LH1 rings is of course a dynamic feature, governed by BChl–BChl interactions and fluctuations of the site energies. Like for the B800 ring, energy transfer within B850 and LH1 has been studied by ultrafast transient absorption,^320–323 fluorescence^299,324 and 2DES³²⁵ measurements of isotropic and anisotropic signals, as well as photon echo methods.³²⁶ A large number of studies have been performed and although the precise numbers vary somewhat between the different measurements there is consensus regarding the main features of the energy transfer and the picture is very similar for both LH2 and LH1. Isolated complexes as well as membrane preparations of several different bacteria have been studied, without pronounced species or preparation type dependence. Following excitation, all studies agree that at room temperature there is a fast 50–100 fs decay of anisotropy and a very similar spectral relaxation (red-shift). In most studies a slower process, ranging from approximately one to a few ps is also reported in both anisotropy and isotropic decays. At low temperature (77 K and 4 K), the femtosecond process remains virtually unchanged, but the slower process becomes slower, particularly in the red part of the spectrum.³²⁰ The energy transfer dynamics of the B850 and LH1 BChl rings have been modeled by several authors, and found to be described by an ∼100 fs energy hopping time in a spectrally inhomogeneous pigment system. It is interesting to note that energy hopping between monomeric BChls was not found to give a proper description of the dynamics;³²⁰ energy hopping between dimeric sites was used in several studies,^299,324 which may be seen as a result of the relaxed exciton size.

Interestingly, a recent anisotropy 2DES study of B850 at 77 K found a temporal mismatch of a few hundred femtoseconds between energy relaxation and depolarization, with the latter substantially slower.³²⁵ It was suggested that energy relaxation can happen locally without spatial motion, which appears to be hindered at cryogenic temperatures, especially at the low energy side. Whereas low mobility was explained by the presence of substantial energetic disorder in the B850 ring, the mechanism of the local relaxation remains to be identified.

Finally, coherent oscillations were also reported for B850 and LH1; at low temperature, a strong oscillation at 105–110 cm⁻¹ and damping time of several hundred femtoseconds were observed, as well as weaker oscillations of higher and lower frequency. The oscillations persisted up to room temperature, albeit with reduced amplitude.^{82,323,324,327} In contrast to similar observations for the FMO protein,⁹⁶ the oscillations were assigned to vibronic coherence without attributing a significant role in light-harvesting function.

In addition to BChl a, other key pigments in purple bacterial light-harvesting complexes are carotenoids. Their contribution to the absorption spectrum of LH1/LH2 fills the gap between the BChl a Soret and Q_x/Q_y bands, pointing to their importance for light harvesting. Indeed, fluorescence excitation spectra reported more than 40 years ago^95,328 provided clear evidence of their role as light-harvesting pigments in LH2 and LH1 complexes. The first attempts to resolve the carotenoid-BChl energy transfer rates were reported by Gillbro et al.,³²⁹ but the data were limited by picosecond time resolution, which, as was shown later, was not sufficient to precisely determine the energy transfer rates. The first application of ultrafast spectroscopy with femtosecond time-resolution to study carotenoid-BChl energy transfer was carried out on the LH2 complex of R. acidophilus, revealing sub-picosecond rates for both Car-B800 and Car-B850 energy transfer.³³⁰ 

These initial studies paved the way to a number of subsequent systematic studies, which, together with significantly improved instrumentation at the end of last century, revealed details about the energy transfer pathways involving carotenoids in purple bacterial antennae. It is well known that purple bacteria accommodate a large variety of carotenoids having conjugation lengths from 9 (e.g., neurosporene in the G1C strain of Rb. sphaeroides) to 13 (spirilloxanthin in Rhodospirillum rubrum). The conjugation length of a carotenoid is the key parameter determining the efficiency and pathways of carotenoid-BChl energy transfer. Reconstitution of LH2 from Rb. spheroides with spheroidene analogs having varying conjugation lengths revealed that prolongation of the conjugation length decreases the efficiency of carotenoid-BChl energy transfer.³³¹ This observation was further corroborated by studies of various LH2's binding different carotenoids. Thus, shorter carotenoids, such as neurosporene, transfer energy with nearly 100% efficiency, while the long ones, such as spirilloxanthin, barely reach 30% efficiency.^151,332

The crucial distinction is the efficiency of energy transfer via the carotenoid S₁ state. While the S₂ pathway has an efficiency in the 25%–40% range for essentially all LH2 and LH1 complexes, the S₁ pathway is active only for carotenoids with the S₁ state energy high enough to allow transfer to the Q_y state of either B800 or B850. The breaking point is between N = 10 and N = 11, as spheroidene (N = 10), has an active S₁ pathway with ∼80% efficiency, while lycopene or rhodopin (N = 11) exhibit essentially no transfer via the S₁ pathway.^332–335 These experimental results were also supported by rapidly developing computational techniques that were able to calculate carotenoid–BChl interaction and to reproduce reasonably the observed energy transfer rates.^336,337

The picture that emerged from a number of studies at the turn of the century was, however, slightly complicated by experimental evidence of other spectral features that were unexplainable within the simple framework of two carotenoid excited states, S₁ and S₂, donating energy to two BChl states, Q_x and Q_y. First, transient absorption experiments extended to the near-IR region revealed a minor depopulation pathway from the excited S₂ state, leading to formation of a carotenoid cation radical.^335,338 The charge separation occurs between carotenoid and B800 BChl a and lasts for a few picoseconds. A possible functional role of this mechanism has never been elucidated. The efficiency of the radical formation is higher for shorter carotenoids, reaching about 10% for neurosporene and is nearly absent for carotenoids with N > 10.³³⁸ Furthermore, the enigmatic carotenoid S^* state (see Sec. III A 2 b) was identified in LH2 and even proposed to serve as a minor energy donor in carotenoid-BChl energy transfer.²¹³ The peculiar dependence of the S^* signal amplitude on excitation intensity led to a proposal of complicated relaxation and energy transfer schemes,²⁰⁵ but it was shown later that the source of the S^* signal in LH2 could be due to an electrochromic shift of the carotenoid absorption band.²¹⁶ The electrochromic shift originates from the local electric field generated by the excited BChl nearby.³³⁹ Niedzwiedzki et al.²¹⁶ showed that if this effect is taken into account, transient absorption spectra of LH2 complexes can be fully explained without invoking the S^* state.

Having examined the intra-complex dynamics of LH2 and LH1, we are now prepared to explore the inter-complex energy transfer dynamics and the flow of energy through the antenna network toward the RC where it is trapped through charge separation, forming a transmembrane potential. Purple bacterial membranes have been imaged with atomic force microscopy at a resolution of ∼10 Å,^340–342 showing LH1-RC complexes surrounded by a sea of LH2 complexes (Fig. 9). From this organization of the PSU, we expect energy transfer first between the LH2 complexes, then from LH2 to LH1, followed by a final transfer step from LH1 to the special pair in the RC. Studies of these processes of course require the presence of the complete PSU, or at least the relevant parts. The experiments have, therefore, been performed on membranes or whole cell preparations of several wild-type or mutant purple bacteria species.

Early pre-LH2-structure picosecond absorption studies reported a LH2-LH1 energy equilibration time of 37 ps (Ref. 293) and, with time-resolved fluorescence, two transfer times to LH1, 10 and 50 ps, were reported for LH2 complexes in close contact and more distant, respectively.³⁴³ Later, in transient absorption experiments with femtosecond resolution a fast ∼5 ps B850 to LH1 energy transfer time was resolved,^297,344 as well as a slower 26 ps component.³⁴⁴ The fast ∼5 ps component was interpreted as energy hopping from a LH2 complex directly associated with a LH1 complex, and the longer 26 ps transfer time was interpreted as energy migration in the LH2 pool prior to transfer to LH1. In relation to the purple bacteria PSU, all these results taken together, suggest that direct LH2-LH1 ring-to-ring transfer occurs with an ∼5 ps transfer time and energy diffusion over the network of LH2 rings prior to hopping over to the LH1 ring takes 30–50 ps. The latter B850 diffusion time could be imagined to depend on the extent of the LH2 peripheral antenna (i.e., number of LH2s per LH1-RC), which probably varies between purple bacterial species and growth conditions.

With an overall energy transfer time of approximately 50 ps from the moment of light absorption in the peripheral antenna until energy is deposited in the LH1 ring of BChls in the vicinity of the RC, light harvesting in purple bacteria up to this point has nearly 100% quantum efficiency. It now remains to see how energy enters the reaction center and initiates charge transfer from the BChl special pair (P₈₇₀) to the monomeric BChl primary electron acceptor. Several time-resolved absorption and fluorescence studies of a few different purple bacterial species, with excitation of the LH1 core antenna, have shown that energy transfer to the special pair occurs with a characteristic time of 40–50 ps.^31,345–348 Knowing that the primary charge separation in the reaction center takes about 3 ps, it was concluded that the overall rate of energy trapping by the RC is limited by LH1 to P₈₇₀ energy transfer, rather than the primary charge separation in the RC.^31,345–348 This was confirmed by studies of purple bacterial membranes of RC mutants having a slowed down charge separation—a wide variation of CT times had only little effect on the overall trapping time.^349,350 Traditionally, two types of trapping kinetics have been identified, “diffusion limited” and trap limited, respectively, where energy diffusion through the antenna network, or charge separation in the RC are rate limiting for the overall trapping rate. Distinct from either of these limiting cases, trapping in purple non-sulfur bacteria was termed “transfer-to-trap-limited.”³⁴⁷ The various energy transfer times characterizing the light-harvesting processes in purple bacteria are summarized in Fig. 10.²⁸¹ 

Based on AFM pictures of the purple bacterial membrane (Fig. 9) and the PSU model in Fig. 10, it can be deduced that the distance from BChls in the LH1 ring to P₈₇₀ is the largest energy donor-acceptor distance in the whole PSU. It was realized that this distance, limited by the size of the RC protein and shape of the LH1 core antenna, is the reason for the relatively slow LH1-P₈₇₀ energy transfer.²⁸¹ It was further speculated that this large distance is necessary to avoid oxidation of LH1 BChls by the oxidized special pair; if this would happen, an oxidized BChl molecule in LH1 would be a very efficient quencher of excitation energy.²⁸¹ The slight uphill energy transfer condition (going from B875 in LH1 to P870 in RC)³⁵¹ most likely serves as a protection mechanism, where energy can be easily detrapped from the closed RC under high light conditions.

What ultrafast spectroscopy taught us about photosynthetic purple bacteria

Ultrafast spectroscopy has given us a detailed picture of the energy flow through the photosynthetic unit of purple bacteria, from the ultrafast, local, inter-chromophore energy transfer within the LH1 and LH2 antenna proteins, to more long-range transfer between the antenna proteins and antenna to RC.

• The peripheral LH2 antenna exhibits energy transfer on two time scales—ultrafast sub-ps migration of excitons delocalized over a few (∼4) BChl molecules within the ring of strongly coupled B850 molecules, and slower, ∼1 ps, transfer from localized B800 excitons to B850, as well as energy migration among the B800 molecules.

• Energy dynamics within the BChls of LH1 is similar to that of B850.

• Ring to ring LH2-LH1 energy transfer occurs with a characteristic time of ∼5 ps, and transfer over several rings takes more time, a few tens of ps.

• The final energy transfer step from the LH1 ring to the RC is the slowest step in the overall light-harvesting process, ∼35 ps.

• Carotenoids collect blue and green light and transfer it with high efficiency on the sub-ps to ps timescale to the BChls of LH2 and LH1.

• All these individual energy transfer steps add up to an overall antenna-to-RC energy transfer time of ∼60 ps for photochemically active RCs, resulting in very efficient conversion of absorbed photons to separated charges in the RC.

In plants, photosynthesis takes place in chloroplasts with the help of chlorophyll and carotenoid molecules organized in two photosystems, photosystem I (PSI) and photosystem II (PSII). The light-harvesting machinery of PSII, situated in the grana membranes of the chloroplasts, consists of several different LHCs feeding excited state energy into the RCs they surround. A PSII supercomplex, called C2S2M2, has been isolated and structurally characterized from several species at various levels of resolution.^352–354 It is a dimer complex, the one from higher plants contains 326 pigment molecules bound to a protein assembly consisting of LHCs surrounding two RCs.³⁵² Each monomer consists of two light-harvesting complexes II (LHCIIs), three minor LHCs (CP24, CP26, and CP29), two core antenna complexes (CP43, CP47), and one RC. Energy transfer has been studied in all the individual, separately isolated, complexes, as well as various membrane preparations, which may be expected to hold multiple copies of interconnected C2S2M2 supercomplexes.³⁵⁵ We will discuss the results for the various levels of organization with the aim to obtain a picture of the energy flow through the PSII light-harvesting network of pigment–protein complexes.

Before we discuss energy transfer in PSII LHCs, we give a short account of their structures. LHCII is the major peripheral light-harvesting complex (Fig. 11), present as a trimer in chloroplasts, binding ∼50% of all chlorophylls in green plants. Its three-dimensional structure was first determined by electron diffraction on two-dimensional crystals, initially to a resolution of 6 Å (Ref. 356) and three years later to 3.4 Å.³⁵⁷ The LHCII structures based on two-dimensional crystals did not have sufficient resolution to distinguish between Chl a and Chl b molecules, or to establish the direction of their x and y axes. However, all this was achieved with the structure obtained later to 2.72 Å resolution with x-ray crystallography,³⁵⁸ providing unambiguous determination of the identities and positions of 8 Chl a and 6 Chl b molecules. In addition, four carotenoid molecules were identified, two luteins and one neoxanthin and violaxanthin each. CP24, CP26, and CP29 are smaller peripheral LHCs present in monomeric form in the photosynthetic membrane (Fig. 11). Early structural models of CP24 indicated binding of five Chl a and five Chl b molecules, as well as two carotenoids, lutein and violaxanthin.³⁵⁹ A more recent high-resolution structure,³⁶⁰ however, shows that the complex binds six Chls a, five Chls b, an