Evading the Illusions: Identification of False Peaks in Micro-Raman Spectroscopy and Guidelines for Scientific Best Practice

: Micro-Raman spectroscopy is an important analytical tool in a large variety of science disciplines. The technique is suitable for both identification of chemical bonds and studying more detailed phenomena like molecular interactions, material strain, crystallinity, defects, and bond formations. Raman scattering has one major weakness however: it is a very low probability process. The weak signals require very sensitive detection systems, which leads to a high probability of picking up signals from origins other than the sample. This complicates the analysis of the results and increases the risk of misinterpreting data. This work provides an overview of the sources of spurious signals occurring in Raman spectra, including photoluminescence, cosmic rays, stray light, artefacts caused by spectrometer components, and signals from other compounds in or surrounding the sample. The origins of these false Raman peaks are explained and means to identify and counteract them are provided.


Introduction
Raman spectroscopy has developed into a widely used and versatile tool in scientific research.It is used in a large variety of disciplines including chemistry, [1] physics, [2,3] biology, [4] geology, [5] gemology, [6] and medicine. [4]In material science, the pharmaceutical industry, and geology it has become a standard tool for research, especially in identification and distribution analysis of molecules, materials, and minerals.Raman spectroscopy can be used for a large number of applications.Apart from chemical identification, it can also be used for detecting material stress and strain, studying the degree of crystallinity, and in situ detection of new chemical bonds forming in a reaction.][9] Three major advantages of the Raman technique are that it is nondestructive if care is taken with the illumination intensity, that no or little sample preparation is needed, and that almost any type of sample can be measured: crystalline, amorphous, liquids, and gases.It is also suitable for measuring directly on or in devices, in biological materials, and for archaeological items and artwork.The exceptions, that cannot be measured, are monoatomic noble gases and most metals and alloys which are Raman inactive due to their electronic structures and symmetry.There is, however, one major weakness with Raman scattering: It is a very low probability process unless probed in resonance with an electronic excitation.Only one photon out of approximately 10 7 is Raman scattered.To detect such a weak signal, a very sensitive detection system is needed which subsequently leads to a high probability of picking up a signal from another origin than the intended measurement.Another consequence is that any small displacement, small defect, or wear of the spectrometer components, lead to changes in the recorded spectra, something that would not be a problem to the same extent in less accurate spectrometers.This is undesirable since all other signal contributions will interfere with, or in worst case completely mask, the weak Raman signal.Some of these signals give rise to Raman peak-like features in the spectra and they are referred to as false Raman peaks.If they are not correctly identified they will complicate the analysis of the results and increase the risk of misinterpretation of the data.Raman spectroscopists have often learned this the hard way and it often leads to extra work, lost time, or that the measurements need to be redone.This work aims to review the different types of signals that can occur in Raman spectra.We present a comprehensive collection of causes of false Raman peaks as well as the means and measures to identify and counteract them.Categorization has been done by the origin of the signal and it is described when and why the false Raman signals occur and how they can be identified and avoided.

The discovery of the Raman effect
The history of Raman spectroscopy starts in 1923 when Smekal, with the use of semi-classical quantum mechanical theory, predicted a new type of inelastic light scattering phenomenon. [10]The Kramer-Heisenberg dispersion formula describing the cross section for scattering of photons by electrons by using the correspondence principle was developed in 1924 and 1925, [11][12][13] and the quantum mechanical derivation by Dirac in 1927 [14,15] put this on a more rigorous foundation.The first experimental evidence came five years after the first prediction when two research groups, Raman and Krishnan, and Landsberg and Mandelstam, independently demonstrated the effect. [16,17]Raman and Krishnan presented their findings first and the scattering effect was named Raman scattering.In 1930 Raman was awarded the Nobel prize in physics for the discovery, where filtered solar light was used as a light source.The journey into becoming a standard spectroscopy technique was rather slow due to the lack of better light sources but changed upon the invention of the laser. [18]When the first laser was built in 1960 [19] the research community immediately realized its usefulness and the first Raman spectrum using a laser was published within 2 years. [20]In the 1970s Raman spectroscopy gained in popularity filling a need to study molecules in aqueous solution and other polar media, where infrared (IR)spectroscopy is hampered by the fact that the media itself has strong IR absorption. [21]Another advantage in Raman spectroscopy is the ability to measure low wavenumber vibrations, not easily attainable in IR spectroscopy.In the 1990s several important technological advances had reached far enough for Raman spectroscopy to benefit from them: The development of affordable lasers, dichroic Rayleigh blocking filters, high-density diffraction gratings, and charge-coupled device (CCD) detectors with improved sensitivity.These components strongly improved the throughput and performance of Raman spectroscopy systems.The typical measurement time was reduced from days to hours and seconds, which improved the usability and advanced Raman spectroscopy from a specialty technique to a readily available standard technique.Raman spectroscopy is now used for scientific research in physics, chemistry, biology, geology, and medicine and has in addition to this also become an established analysis technique in gemology, the semiconductor industry, pharmaceutical development, process monitoring, and as a healthcare diagnosis tool.There are now more than 25 subdisciplines of Raman [21] out of which resonance Raman, surface enhanced Raman scattering (SERS), and coherent anti-Stokes Raman scattering (CARS) are some of the most known.In this work, the focus is on conventional confocal Raman spectroscopy while the other varieties of Raman spectroscopy are briefly explained for comparison.For recent advances and an overview of more specialized techniques we refer to other sources. [1,21]

Raman scattering principles and theory
When light interacts with matter, a fraction of the light can be scattered.The scattering can be elastic, where the energy of the photon is unchanged by the event, or it can be inelastic where the energies of the incident and exiting photons differ.For molecules and crystal lattices where the scattering centers are smaller than the wavelength of the light, the elastic Rayleigh scattering is the dominating process, resulting in a strong emission of light with the same wavelength as the incident light.A small fraction of the scattered light, about one out of 10 7 photons, is instead inelastically scattered.The incoming photon interacts with the electron clouds in a molecule or crystal lattice and under conditions clarified below, energy can be transferred from the photon field to a vibrational bond or from the vibrational bond to the photon field.The photon can thus either lose or gain energy quanta corresponding to the energy difference in-between vibrational energy levels in the bond, where the origin for this will be explained below.The selection rule that determines if a bond vibration is Raman active states that the bond needs to change the polarizability of the electron cloud during the vibration.In group theory symmetry analysis this corresponds to the modes with quadratic functions.The polarizability ellipsoid can be changed in three ways: in size, in shape, or in orientation.
Similar selection rules exist for IR absorption: a vibration is only IR active if the vibration changes the dipole moment of the chemical bond.This makes the IR and Raman spectroscopy techniques complementary, especially for molecules that have a center of inversion, since vibrations in such molecules cannot be both Raman and IR active.We can use the classical electromagnetic theory approach to explain the phenomenon of laser-induced Raman scattering.The incident photon (from the laser) can be described by the time-dependent electromagnetic field strength from its frequency-dependent electromagnetic field, E=E 0 cos(ω 0 t), with incident photon energy, and its angular frequency are denoted E and ω 0 , respectively.The electromagnetic field from the laser can induce a dipole moment μ in electron clouds around atoms, where the resistance to induce a dipole with the applied field can be expressed as μ = αE(ω 0 ) where α is the polarizability of the electron cloud.
If we now consider a diatomic molecule and that the ability to polarize the electron cloud around these two atoms can change when they are closer or further away from each other in a molecular vibration, one can formulate this as a change of the polarizability with α = α 0 + (δα/δq)q 0 for a displacement with respect to a coordinate q.In a more general expression, one can thus expand α in a Taylor series and differentiate with respect to the normal coordinates Q k of a vibration mode to have: Where α 0 is the polarizability at the equilibrium position and α' k is the derivative of the polarizability with respect to the normal mode k, which describes the change in polarizability during the vibration.Here, we can assume that the time dependence of the oscillating dipole follows a harmonic motion and its time dependence can be written as where Q k 0 is the amplitude of the normal coordinates, ω k is the vibrational frequency of the mode, and δ k is a phase factor.Inserting into equation 1 gives: One can now formulate the induced dipole in an equation with the time-dependent incoming field and a polarizability that includes a change when displaced, as described in equation 2, with instantaneous response and a phase factor δ k = 0, to yield: The ability to induce a dipole in the time-dependent field is now described by the polarizability at the equilibrium bond length plus a contribution that depends on the change in polarizability upon vibration.The latter is a coupled oscillator in between the incoming field oscillating with the angular frequency ω 0 and the molecular vibration ω k along mode k in the molecule.At first sight, this looks difficult to assess, but the coupled oscillation can be reformulated using the trigonometric identity cos a The first term in equation 4 contributes to the Rayleigh scattering; its frequency is the same as the incident light and not dependent on the frequency of molecular vibrations.The second term gives the dipolar contribution to the inelastically scattered light with a gain in frequency (anti-Stokes) and the third term to the loss in frequency (Stokes).Its frequency is dependent on the incident light frequency and the molecular vibration, and its two frequencies will be located symmetrically around the incident light frequency.Notably, if there is no change in polarizability during the vibration; that is, if α ' k = 0, the term for inelastic scattering will vanish and the mode will thus not be Raman active.To use this in the context of the scattered field, one can start by considering that an accelerated charge emits radiation, and the second derivative of the dipole enters as a key part in the re-radiated field E R [22] and can be expressed, upon a displacement sinθ, with: Inserting the second derivative of the dipole contribution describing the loss or gain frequency in equation 4 into equation 5, the intensity of the Raman scattering I Raman = I s / I 0 can be formulated as: Where E 0 is the incoming field, E R the re-radiated field, and ω s is the Raman scattering frequency ω 0 + ω k (or ω 0 -ω k ).Noteworthy in this equation is the strong intensity dependence on scattering frequency, to the power of four and thus 1/λ 4 with respect to the wavelength of the incoming light.The description in equation 6 is dependent on a fixed polar coordinate system and radiation at point r, θ, in the plane of φ.In a simplified notation, without the inclusion of the normal coordinates Q k of a vibration mode and using the instantaneous dipole μ = μ 0 cos(ωt), with a magnitude of μ 0 and performing a time-averaging of the scattering, one arrives at: [23] I Raman ¼ w 4 s m 2 0 sin 2 q 32p 2 e 0 c 3 0 (7)   Again there is a ω s 4 dependence, but here with the average magnitude of the dipole moment instead of the time-dependent dipole oscillation.The strong dependence of the frequency is a motivation for using excitation lasers with short wavelengths to increase the signal intensity of the scattered light.For the ideal case obeying equation 7, the scattering intensity difference between a 405 nm and 1064 nm laser utilized on the same sample (assuming the same Raman scattering cross section) is 1/405 4 : 1/1064 4 is 48, and thus, 48 times longer measurements are needed to extract the same scattering intensity under non-resonance conditions.A 1 h measurement for a certain desired peak intensity using a 405 nm laser would thus translate to a 2 day (48 h) measurement using a 1064 nm laser.The Raman scattering frequency ω s and the incident light frequency ω 0 are not equal for Raman scattering but, with regards to scattering intensity, they are very close.The classical (nonquantized) picture of light-matter interaction can explain the origin of the number of vibration lines and their position.Although this is very useful and gives an intuitive understanding of the Raman process, the classical view cannot explain the appearance of rotational Raman activity, as a quantum mechanical view is required to obtain quantized rotation levels.Furthermore, it does not provide a connection between the polarizability and the transition dipole moment with the molecular Hamiltonian and frequency of the incident field, as is important in resonance Raman scattering.A quantum mechanical description can instead be formulated by using a second-order perturbation theory first applied by Kramer and Heisenberg [13] and Dirac [15] where they derived a quantum mechanical description of the molecular polarizability α ρσ with: Where j i > , j r > , and j f > refer to the initial, intermediate, and final states of the Raman scattering process, ρ and σ denote the polarization of the incident and scattered light, while ω 0 is the laser frequency, and ω r i and ω r f denote the energetic differences between the respective states.The damping constant Γ has been introduced to incorporate the homogeneous line widths of the respective band and is related to the inversed lifetime of the intermediate state.The first term in equation 8 represents absorption of a photon into an intermediate, scattering state, followed by emission, of the inelastically scattered Stokes photon.The second term instead represents the reversed order of events.If the excitation wavelength is chosen resonantly with an electronic transition, the denominator of the first term in equation 8 becomes small and the term will dominantly contribute to the polarizability and thus, largely, the observed Raman scattering.
The transitions involved in Raman scattering can be illustrated using schematic quantum level energy diagrams called Jablonski diagrams where the incident and scattered photons are represented by arrows.In the left part of Figure 1, Jablonski diagrams for anti-Stokes Raman, Rayleigh, and Stokes Raman scattering are shown together with a spectrum that exhibits the corresponding processes.As seen in Figure 1 energy differences between the vibrational states in the molecule are relatively small compared to the energy differences between electronic states.The incident light, represented by a green arrow, excites the molecule to an intermediate state, referred to as a virtual state.This virtual state is not a quantum mechanically allowed energy state and the molecule, within a lifetime proportional to 1/Γ, relaxes to an allowed state and emits a photon that corresponds to the difference in energy between the virtual state and the final state.If the initial and final states are the same, the wavelength of the emitted light will be equal to that of the incident light, which corresponds to Rayleigh scattering.About 1 out of 10 3 -10 4 photons are Rayleigh scattered so it is a much higher probability process than Raman scattering.For the Raman scattered light there are, as derived above, two cases.One case is that a molecule at ground state can be transferred to a higher vibrational state which results in scattered light with lower energy.This red shifting of the light is named Stokes Raman scattering.The other case is that a molecule that is already at a higher vibrational state is transferred to the ground state which results in scattered light with higher energy.This blue shifting of the light is named anti-Stokes Raman scattering.
The Raman scattering processes are named after Stokes law of fluorescence that states that fluorescent light must have equal or lower energy than the incident light.Raman scattering is a different mechanism than fluorescence, but Stokes Raman fulfills Stokes law while anti-Stokes Raman does not.In Figure 1 the processes of Rayleigh, anti-Stokes Raman, and Stokes Raman are shown together with a Raman spectrum of TiO 2 (anatase) exhibiting peaks from these three processes.At room temperature most of the molecules with low molecular weight will be in their vibrational ground state.Therefore, there is a higher probability for a photon to interact with a ground state molecule and this explains why Stokes Raman usually has a stronger intensity than anti-Stokes Raman.The populations of the different states follow the Boltzmann distribution and are dependent on both temperature and the energy difference between the states.The ratio of higher to lower state fraction can be expressed by N 1 /N 0 = exp(À hcν/k B T), where N 1 is the population of the higher state, N 0 is the population of the lower state, h is Planck's constant, c is the speed of light in ms À 1 , v is the Raman shift in m À 1 , k B is the Boltzmann constant, and T is the temperature in Kelvin.From this expression one can see that the probability of the molecule or material being in a higher vibrational state increases significantly in materials that have vibrations with a small Raman shift, such as vibrations including heavy atoms, and a corresponding low energy difference between the ground state and the excited vibrational state.Comparing an I 2 molecule (Raman shift 180 cm À 1 ) with H 2 (Raman shift 4156 cm À 1 ) at room temperature yields anti-Stokes fractions of 0.41 and 1.4 • 10 À 7 respectively, the large difference being entirely due to the energy difference between the vibrational states.Materials or molecules at elevated temperatures also exhibit stronger anti-Stokes Raman peaks from thermal excitation to higher vibrational states.The ratio of anti-Stokes and Stokes signals can in fact be used to determine the temperature of the sample, which is sometimes used in analysis of combustion gases.There are several subdisciplines of Raman spectroscopy that can obtain an increased scattering intensity compared to classical Raman scattering.In the right part of Figure 1, the principles for stimulated Raman scattering (SRS), coherent anti-Stokes Raman scattering (CARS), resonance Raman scattering, hyper Raman scattering, surface-enhanced Raman scattering (SERS), and tip-enhanced Raman scattering (TERS) are displayed.SRS use two lasers, a pump laser, and a Stokes laser.Notably, if the pump laser is considered an external field, SRS can be described as a two-photon process, but as a four-photon process if the pump laser is included in the internal process.The Stokes laser wavelength is either scanned across the wavelength range or tuned to match a specific vibration.When the difference between the pump and Stokes lasers match a vibration in the sample, the transition probability increases and the emission from the sample at the Stokes wavelength increases, whereas the light emission at the pump laser frequency decreases.The measured parameter in this technique is therefore not a frequency shift, as in most other Raman techniques, but a change in the ratio of pump and Stokes wavelength intensity, a delta emission.CARS is based on a three-photon process.It uses one laser that acts as both a pump and a probe and one Stokes laser.The difference between the pump and Stokes laser is tuned to match a specific vibrational bond.This will populate the chosen higher vibrational state and the molecules are excited by the probe laser; as they return to the ground state, anti-Stokes photons are emitted.In resonance Raman, the wavelength (and energy) of the excitation light matches an electronically excited state of the sample.The scattering probability is strongly enhanced due to the allowed intermediate quantum state.The excitation in Hyper Raman is a two-photon process where the vibration is simultaneously excited by two photons from the same laser.This process is weaker than conventional Raman and typically requires a much higher laser intensity or the simultaneous utilization of resonance Raman or SERS but has the advantage that the double excitation breaks the Raman selection rules, which makes it possible to see non-Raman active modes (IR-active and silent modes).In SERS a signal enhancement is achieved by the near-field and charge transfer in the proximity of the surface of plasmonic nanoparticles.The nanoparticles are usually noble metals (e.g., Au or Ag), and can be deposited on a substrate or added as free particles to a sample in solution.The particles are surrounded by a plasmonic field, which enhances the scattering of the molecules.The scattering is dependent on the element and size of the plasmonic particle as well as on the wavelength of the incident light, and the enhancement factor can be as large as 10 9 . [24]There is no special instrumentation needed to perform SERS measurements; the acquisition is done with a conventional Raman spectroscopy setup.TERS is very similar to SERS but in TERS a noble-metal-coated (Au or Ag) AFM tip or a plasmonic particle attached to an AFM tip is put in close proximity to a sample surface.This enables a spectral intensity enhancement that is localized around the AFM tip, which gives a high spatial resolution.

Scientific Perspective
Raman spectra are usually presented as Raman shift.The energy of the excitation laser is set to zero and the energy shift compared to the laser is presented in reciprocal centimeters (cm À 1 ).By convention, Stokes Raman are presented as a positive shift, mainly because Stokes Raman is much more commonly used and it would be inconvenient to always present data in negative numbers.

Interpretation of Raman data
As it has been shown above, each Raman peak is related to a specific vibration and the vibrations are in turn related to chemical bonds.Chemical bonds that are adjacent can, however, combine and give rise to several different vibrations (symmetric, asymmetric, and bending).The complexity of the Raman spectra therefore grows fast with the size of the molecules.The link between the Raman peaks and bonds can give some intuitive understanding for interpretation of Raman spectra.Two identical molecules have the same bonds and their spectra are also identical, enabling Raman to be used for compound identification.This is arguably the most common use of Raman spectroscopy.Expanding this reasoning and considering a chemical reaction that creates a new type of bond in the molecule, this will lead to the occurrence of a new Raman peak in the spectrum or to an increase in intensity of an existing peak, and breaking of bonds or removal of functional groups will reduce the number of peaks or reduce the intensity of an existing peak.If we look at a crystalline material, the Raman signal originates from both local modes and collective vibrations in the crystal lattice.For crystals that contain defects there could be both a local broken symmetry and new bands as well as slight variations in bond lengths throughout the crystal, which could be seen as broadening of the Raman peaks.This means that the occurrence of new peaks can be used to analyze the type of local defects and the peak widths can be an indicator of degree of crystallinity.Raman can also distinguish between different crystalline phases of the same compound (e.g., anatase, brookite, and rutile TiO 2 ) since the bond arrangements differ.For oriented single crystals with a non-identical unit cell axis and direction specific bonds, Raman Spectroscopy can be used to determine the orientation of the crystal, since the laser light only excites the bonds that match the laser polarization direction and not bonds perpendicular to this.In a material subjected to stress, the bond lengths will be different than in a relaxed material, which will lead to a shift of the Raman peak position.For compressive stress the bond lengths will be shorter, corresponding to a higher vibration frequency and a higher Raman shift, and for tensile stress the bonds will be longer with a resulting lower Raman shift.Shifts due to material stress are usually small, within a few cm À 1 of the equilibrium position.As a support for the interpretation and identification of Raman peaks, group theoretical analysis or density functional theory (DFT) calculations can be used.DFT will numerically calculate the Raman active fundamental vibrational modes of molecules and crystals and their vibration frequencies, where many implementations of DFT also include group theoretical considerations.[29] After this introduction of Raman scattering we turn our attention to practical Raman spectroscopy measurements and the spectral artefacts that can occur in Raman spectra.This section is specifically dedicated to confocal micro-Raman spectroscopy, which is most commonly used in scientific research, but several of the phenomena and artefacts described are equally relevant for transmission Raman spectroscopy or smaller handheld devices.

Cosmic rays
One of the most common causes of spectral artefacts is naturally occurring cosmic rays.Primary cosmic rays refer to high energy charged particles (protons, electrons, and atomic nuclei) that reach earth from space. [30]They can be divided into two categories: galactic cosmic rays and solar cosmic rays, also called solar wind.The former originates from distant supernova explosions and mainly consists of atomic nuclei, while the latter mainly consists of lighter particles such as electrons, protons, and alpha particles, which give rise to the aurora light phenomenon.Due to their electric charge the particles are deflected by the earth's magnetic field and are hence more common at high latitudes.A heavy primary cosmic ray that collides with atoms in the earth's atmosphere causes a hadronic cascade that produces secondary cosmic rays.At ground level 98 % of the cosmic rays consist of muons, which are heavy charged particles that travel through any instrument enclosure or shielding and directly hit the spectrometer detector, at a random position, without going through the microscope or the spectrometer. [31]Muons travel in a straight path and, unless they come in exactly parallel to the detector surface, they only hit one or two pixels on the CCD, creating very sharp peaks.They can in many cases give a very high signal response.In Figure 2, two different manifestations of cosmic rays are shown: the dim green light of an aurora borealis caused by solar cosmic rays and sharp peaks in a Raman spectrum caused by galactic cosmic rays.
Since cosmic rays are random in both time and space they will not occur in the same position if the measurement is repeated.This means that multiple measurements or remeasuring procedures are suitable actions to identify cosmic rays in the spectra.It is common that methods for identifying and/or deleting cosmic rays are included in spectroscopy software.The sharpness of the peak is a strong indication that the peak is unnatural, but there are also other possible causes for such sharp peaks, as will be discussed later.

Stray lights and insufficient shielding
If Raman spectra are acquired in a micro-Raman system that is not enclosed (i.e., a custom-built system on an optical table or a system with an open microscope), one is likely to encounter problems with stray light from light sources in the room.Fluorescent lights from the room illumination, computer screens, and LEDs on various electronics fall under this category.The positions of peaks originating from stray light are independent of laser wavelength and sample and will always occur in the same spectral position if an absolute energy scale (nm, absolute wavenumbers, eV) is used.Measurement with the laser turned off is an appropriate procedure for identification since this removes all true Raman signals.A peak that persists is very likely to originate from stray light.For this category, turning off the room lights and shielding of the experiment from any remaining lights in the laboratory are the recommended actions.This can be done using an enclosure (Figure 3c) or blackout curtains, but can for small samples also be as simple as placing a small cylinder around the sample extending up to the microscope objective, as seen in Figure 3a and 3b.In addition to shielding off the measurement from stray light, such measures will also protect the user from accidental laser exposure.If a light leak into an enclosed system is suspected, gaps can be sealed off with metal tape.

Fluorescent lights
A spectrum of a fluorescent lamp present in our laboratory facilities is shown in Figure 4.The spectrum consists of both sharp peaks from atomic emission lines originating from a plasma and of broader features from a fluorescent coating.It should be noted that there are peaks in the whole visible range and that they also extend into the ultraviolet (UV) range.The sharpest peaks will immediately cause suspicion to the trained eye, but some peaks have a Gaussian or Lorentzian distribution with a reasonable width, just like true Raman peaks.
Fluorescent light peak positions in nm can be converted into Raman shift and vice versa using the following equations: l ¼ 1 1 Where Δω is the Raman shift in reciprocal centimeters, λ ex is the laser excitation wavelength in nanometers, and λ is the peak position in nanometers.This unit conversion can be used when unknown peaks from Raman spectra need to be compared with fixed wavelength peak references like the  spectra in Figure 4 or the laser emission peaks in Table 1.In Figure 4b and 4c, the spectrum of fluorescent lights are shown in Raman shifts when measured with 532 nm and 785 nm lasers, respectively.
be misinterpreted for sample fluorescence, which will be discussed later.Figure 5 (lower) shows a LED-type spectrum.The light emission from tungsten lamps is even broader and will be seen as an increased baseline level.

Computer screens
Computers are necessary parts of data acquisition systems and therefore often in close proximity to the spectrometer setup.Computer screens belong to a diverse category containing several different technologies and numerous designs.The spectral emissions in Figure 5 shows that the first screen uses a glow discharge technology, and the second screen uses an LED technology for the back illumination.
As with room lights, narrow peaks cause more problems than broad spectral features and the recommendation is to measure the spectrum of your computer screen and, as for other sources of stray light, shield off the experiment.

Spectrometer components
There are sometimes faulty components and wear in a Raman system, where optical components could deteriorate over time and perform less well.This may cause disturbances or artefacts in the data acquired.An illustration of the optical components in a lens-based Micro-Raman spectroscopy system is shown in Figure 6a.

Extra laser emission lines
There are two main types of lasers: gas lasers and solid-state lasers.For gas lasers the light is generated in a plasma in a gas-filled cavity.The emitted light originates from atomic emission lines.A gas plasma will have several emission lines and the unwanted plasma lines needs to be filtered away using a clean-up filter.For solid state lasers there is no plasma, instead the light is emitted from a diode junction.Solid-state lasers can still have extra emission lines,  especially if it is a frequency-doubled or -tripled laser, and they also require clean-up filters.If the filter is not removing the extra emission lines completely, these will show up in the spectrum.Figure 6b shows this phenomenon for a 532 nm neodymium-doped yttrium aluminum garnet (Nd: YAG) diode laser: an extra emission line is occurring at the base frequency 1064 nm.The 532 nm laser line is strongly attenuated by the Rayleigh rejection filter whereas the base frequency is not.Due to the narrow emission of Raman lasers, these peaks are very sharp but, in contrast to cosmic rays, they are originating from within the spectrometer system and therefore reoccurring in the same position also with repeated measurements.
Identification is done by measuring on a pure non-Raman-active metal surface.The metallic surface will reflect the laser beam, the excitation wavelength is filtered away by the Rayleigh blocking filter, and any remaining very sharp peaks are then likely to originate from the laser.This should, however, be confirmed by comparing the peak positions with known laser emission lines like the ones tabulated in Table 1 since signals from other spectrometer components can also be detected in such a measurement.A better performing clean-up filter will remove this spectral artefact, but it is often enough to just identify the origin of the peaks.

Gratings
The purpose of the grating is to diffract the light of different wavelengths.The most common gratings are very sensitive and cannot be cleaned in a good way.Therefore, it is crucial to keep grease from fingers and dirt away from them.If the grating is not performing up to standard, a small fraction of the light could be reflected or diffusely scattered instead of diffracted.This will cause an increased light intensity on the center axis of the spectrometer leading to a broad signal near the center of the spectrum, as seen in Figure 7.A condition that needs to be fulfilled for this to occur is that there needs to be a lot of light going through the system (e.g., a large fluorescence background) since only a small fraction of the total light will be diffusely scattered.The signal will be rather broad and if the grating center position is changed to move the acquisition range, the peak will still occur at the center of the range, moving to a new Raman shift position.There is no quick workaround solution to this problem, thus, replacement of the grating would be the action to take.

Insufficient Rayleigh blocking
It is common that Rayleigh blocking filters do not block all the Rayleigh scattered light.As long as the Rayleigh scattered light is in the same magnitude as the Raman scattered light or less, this is not a problem.But if the laser blocking capability of the Rayleigh filter is too low, intense Rayleigh scattered light will saturate the CCD and charge will overflow into neighboring pixels, causing an increased signal or apparent peaks close to the laser wavelength.A warning sign is an increasing background towards low Raman shifts.Acquiring a spectrum where the wavelength range includes the laser wavelength will help to determine whether laser saturation is the cause.Such a spectrum clearly shows if the Rayleigh filter is performing according to standard or if laser saturation is occurring.Depending on the type of CCD, saturation can be displayed as the maximum count number of the CCD, as in the case shown in Figure 8, or as zero counts.If the laser saturates the CCD, it is likely that the Rayleigh filter needs to be replaced or, if the system geometry allows it, one can adjust the angle of the Rayleigh filter to increase the laser blocking to a certain extent.However, this will also affect the alignment and Rayleigh filter cut-off threshold.Note that measurements far from the laser wavelength, at high Raman shift, could still be ok.

Ripples and transmission of filter
The Rayleigh rejection filter could have a certain shape in its transmission curve.It is not always 0 % or 100 %.There is a transition range with a slope in the transmission curve.Ideally this slope is steep, but that is not always the case and there can also be a multistep transition or smaller periodical features (i.e., ripples) caused by interference that can be interpreted as very weak peaks.The signal will not have a Gaussian/Lorentzian shape.These types of spectral features most often occur at low Raman shift close to the filter edge, which coincides with a region where inorganic materials with heavy elements commonly have Raman peaks.The same features will be present in every spectrum taken with that specific filter but, since these are wavelength-dependent differences in filter transmission, they will only be seen when there is a certain amount of light going through the filter, such as a background signal.By introducing broadband white light at the sample position, it is easy to see the shape of the Rayleigh filter; an example of a Rayleigh edge filter exhibiting both a 2-step edge and ripples can be seen in Figure 9.

Pixel defects of CCDs
The most common detector technology in Raman systems are CCDs.In a CCD-detector there are sometimes defective pixels.They can be hot pixels, also called sparklers, that give much higher signal than what they should, cold pixels that give less signal than what they should, or they can be dead pixels that give no signal at all.Hot pixels give sharp peaks and cold and dead pixels give sharp dips in the spectrum.These features appear at the same relative position in every spectrum regardless of which range the spectrometer measures.Defective pixels can be avoided by limiting the area on the detector chip that is used for data collection; however, if they are centrally located this limitation could be difficult or have a large impact on the system performance.

Etaloning
When the light passes through a window or layer with a thickness similar to the wavelength etaloning can occur.It is an interference phenomenon and is seen as periodical fringes that are overlaid on the Raman spectrum.Etaloning occurs more commonly in back-illuminated CCDs, where long wavelength light penetrates through the photosensitive layer and is reflected at its boundaries. [39]Using a shorter wavelength excitation will reduce the risk of CCD-etaloning.To compensate for CCD-induced artefacts like etaloning or pixel defects it is possible to measure on a Raman inactive surface that is reflective (e.g., stainless steel) using the same spectral range as for the sample spectrum and then subtract the steel spectrum from the sample spectrum.Errors caused by the detector will then be present in both spectra and cancel each other.

Photoluminescence
Photoluminescence (PL) is a light-emitting process that occurs in samples and is often seen in Raman spectra.The term luminescence refers to light coming from the sample, while "photo" refers to the process causing light emission.PL is not a scattering process; it is absorption and reemission of photons.The molecules commonly lose some of their energy due to non-radiative thermal processes and the reemitted photons then have lower energies.PL can occur as fluorescence or as phosphorescence.PL emission is often broad, indicating that several similar transitions overlap.Fluorescence is the most common PL process, and it is emitted from an allowed singlet-to-singlet transition and occurs on a time scale of 10 À 10 -10 À 7 s, which is much slower compared to Raman events that occur on a timescale of 10 À 12 -10 À 14 s.The mechanism for phosphorescence is very similar with the important difference being that the light emitted originates from a forbidden transition (triplet-tosinglet).In a phosphorescent material, the molecules in a higher energy state are accumulated during the excitation after which they slowly revert to the ground state and emitting photons, causing a weak long-lasting glow.
In a similar manner, processes from chemiluminescence, thermoluminescence, piezo luminescence, or electroluminescence also emit light and will thus also be visible and detectable in a Raman setup.A Raman spectrometer is, in principle, a highly resolved PL spectrometer that can also detect weak intensity peaks from inelastically scattered light.The difference from a conventional PL spectrometer is that the excitation laser line width needs to be very narrow (ca.0.01 nm), a laser rejection filter that cuts close to the laser line is required and with an optical density of about 6, and the spectrometer needs high enough spectral resolution to resolve vibrational levels.A Raman spectrometer can always measure PL, while the opposite is typically not the case due to the wider linewidth of the excitation source and less accurate spectrometer in conventional PL setups.An optical spectrum that shows contributions from both Rayleigh scattering, Raman scattering, and fluorescence from Cu 2 O is shown in Figure 10.Fluorescence often causes problems in Raman spectroscopy.The strong fluorescence signal is often very dominant, obscuring the Raman signal.The different processes that give rise to Raman scattering and fluorescence can, however, indicate how this can be avoided.Firstly, the fluorescence is dependent on certain closely spaced energy levels being populated.This requires the excitation photons to be of a certain matching energy.By changing the wavelength of the excitation source so that the photon energy does not match the energy states of the molecule or crystal, it is in many cases possible to avoid the fluorescence, and this is the main reason why many research Raman systems are equipped with multiple laser wavelengths.
A general rule of thumb is that organic and biological samples often give less fluorescence when measured with a near-IR laser (e.g., 785 nm), whereas inorganic molecules tend to emit fluorescence at longer wavelengths (Figure 10) and therefore benefit from shorter excitation wavelengths, but there are exceptions to this rule.Secondly, photons emitted by fluorescence have, according to the abovementioned Stokes law, lower energy than the excitation light.Anti-Stokes Raman will therefore not be affected by fluorescence unless the peaks are at a very small Raman shift.Thirdly, the difference in the timescale of the Raman and fluorescence processes makes it possible to separate them by time-resolved spectroscopy.More approaches of fluorescence suppression can be found in a review paper by Wei et.al., [40] but some of them require instrumentation that mainstream micro-Raman systems lack.While PL in the sections above have mainly been described as a competing process and a problem that should be avoided, it is important to point out that the PL signal also gives valuable information about the sample and the energy levels and transitions present.PL spectroscopy is a commonly used analysis technique and, as mentioned above, the large similarity between the optical components in Raman spectrometers and those used in PL spectrometers mean that the Raman spectrometer can be considered a high-resolution PL spectrometer and be used for PL spectroscopy as well.

Sample phase transformation
One of the errors that is most difficult to notice is when the Raman measurement itself creates the illusion.This can happen when the laser used for the measurement induces a phase transition within the sample.The excitation lasers used in modern research grade Raman spectroscopy systems are often in the order of magnitude of 10-100 mW, enabling short acquisition times and high-speed mapping.Commonly neutral density filters are utilized to adjust the light intensity, allowing suppression of the intensity up to a factor of 10 6 , but this will also prolong the measurement with the corresponding factor.It is thus desired to acquire the Raman spectra at as high an intensity as possible without affecting the sample.But at high intensities, the heat transferred to the sample can in some cases induce a phase transition or even burn the sample.If the sample is in air, under-oxidized samples may oxidize.If the sample is combusted, the mistake is soon noticed but a subtle phase transition is easily overlooked.A good example of this is the iron oxide magnetite for which several works have published the Raman spectrum of maghemite or hematite instead of the intended magnetite, most likely due to laser-induced phase transition during the measurement. [41,42]To show, without doubt, that laser-induced phase transition easily occurs in iron oxides we designed an experiment where a magnetite nanopowder was measured with a 532 nm laser.Throughout this experiment the signal was acquired from the same spot on the sample while the laser power was gradually increased for each acquisition, from an initial 0.04 mW to a final intensity of 20 mW.The result is presented in Figure 11 and shows a transformation first into maghemite and subsequently to hematite.To overcome the problem with laser heating the measurements need to be performed with a low laser intensity until the peaks present have been identified.Once the peak positions are known, the laser intensity can be increased to get better statistics, as long as it is established that the intensity is low enough for the spectral signature not to change.Please note that if a powder sample is inhomogeneous, a laser intensity can be low enough for appropriate Raman measurements in one part of the sample, but too high in other parts, depending on the thickness and packing of the powder.Compacting the sample and placing the powder on a heat-conducting substrate can also improve the heat dissipation.The use of excessive laser power can, in some cases, have an influence on the Raman spectrum even if the sample does not undergo a phase transition.The heat delivered by the laser can lead to a local thermal expansion which induces a peak shift. [43,44]

Raman signal from other sources
The Raman peaks in this category are not false, but true and they originate from sources other than the sample.The peaks are due to their Raman appearance commonly misinterpreted as sample peaks and often cause confusion or difficulties when the results should be interpreted.It is important to be aware of the geometry of both the experimental setup and the sample.The sampling volume in a measurement is dependent on three parameters: the laser wavelength, the numerical aperture (NA) of the objective lens, and the slits or the pin holes in the spectrometer set up.Typically, with high confocal spectrometer settings and an objective with high NA, like 50x or 100x, the majority of the signal is collected from within a micron-sized spot on a sample surface.In the z-direction the resolution is usually worse, approximately half.Some signal will still come from the surroundings.It should be noted that the collection volume is at least one order of magnitude larger than the thickness of a 50-100 nm thin film.When focusing in a liquid through a cover glass it is also difficult to determine how deep into the sample the focus point is.Using low magnification objectives with low NA gives less control over the sampling volume.Figure 12 illustrates two measurement situations where signal will be obtained from several species.In Figure 12a there is sample, solvent, and the glass slide.Focusing higher up could avoid the glass slide but would instead risk that the cover glass is included.In case Figure 12b, the sampling volumes include a thin film sample, the substrate, and the air above the sample.

Glass
The inert and transparent properties of glass make it suitable for many measurement setups.It can be used as a sample slide, sample vessel, substrate, cover glass, or optical access window.Glass has an unordered amorphous structure, which results in broad Raman peaks, usually in the range between 400-1400 cm À 1 Raman shift.There are many different types of glass, and it is good practice to measure the spectrum of the glass itself in advance to prevent that the glass signal is mistaken for sample signal.Three examples of Raman spectra from glass are shown in Figure 13.In rare cases the spectrometer can even pick up signal from the glass in the objective lens.This is seen in Figure 14a as the broad peaks at 780 and 980 cm À 1

Air and gasses
In the Introduction it was mentioned that one of the advantages of Raman spectroscopy is that gaseous samples can be measured.Since the sample is most often surrounded by air there is subsequently a possibility to pick up signal from oxygen and nitrogen.Oxygen has a sharp peak at 1555 cm À 1 and nitrogen has a sharp peak at 2329 cm À 1 , which can be seen in Figure 14a.The ratios of the peaks are roughly the same as the ratios of the amounts of O 2 and N 2 in air where their relative concentration can be more accurately determined if one corrects for the slightly different number of electrons, and thus polarizability, in the two gases.For natural reasons, these two peaks usually occur together and are therefore easily identified.The Raman cross section for oxygen and nitrogen gases is relatively low, which results in a low signal intensity, and subsequently long exposure times are needed for these peaks to be clearly seen.If another gas surrounds the sample, that gas will also give a signal, except for monoatomic noble gases that naturally have no molecular vibrations.This highlights the option to use a noble gas atmosphere to exclude influence from the air when designing a measurement setup.Molecules in the gas phase can rotate freely, which give additional degrees of freedom and give rise to rotational quantum levels.Rotational Raman occurs as multiple, periodical, closely spaced (few wavenumbers) peaks.They are present both around the Rayleigh line and around each vibrational peak, but for the latter case they are extremely weak.The peaks around the Rayleigh line are, however, comparably strong, which can be seen in Figure 14b.They abate at around 150 cm À 1 Raman shift and could interfere when measuring low wavenumber peaks.

Substrates
In material science the substrate used is of great importance.In many cases the main purpose of the substrate is to act as mechanical support for the sample, but substrates can also perform other functions like electrical contacting, heat conduction, or act as a growth template for epitaxial growth.
In the case that the sample is not dependent on the use of a specific substrate, it is recommended to use a substrate with Raman signals with low intensity.Metal, metal-coated glass, quartz, fused silica, CaF 2 , and BaF 2 have been proposed for this purpose. [45]In Figure 15 we have included three examples of Raman active metals: Mg, Zn, and Bi, and Raman spectra of a selection of commonly used substrates are shown in Figure 16; the latter include Si, quartz, fused silica, sapphire, polyethylene terephthalate (PET), and stainless steel.For glass substrates we refer to Figure 13.Spectra for all common substrates cannot be included in this paper, instead it is emphasized that the bare substrate should always be measured separately so that its spectral fingerprint is known.This advice also applies to metal substrates and there are two reasons for this.The first is to verify that no Raman active metal oxide has formed on the surface.Oxides can form spontaneously over time or as a result of laser-induced heating during the measurement if a high laser power is used.The second reason is that, in contrast to what many literature sources claim, not all metals are Raman inactive.Metals that crystallize in a hexagonal close-packed structure with two atoms in the primitive unit cell, have optical phonons and can show Raman activity. [46][49] In Figure 15 we have included three examples of Raman-active metals: Mg, Zn, and Bi.Some substrate materials have a few well-defined Raman peaks, as in the case of Si, quartz, and sapphire, some have more broad Raman bands, like glass, and some have a large number of peaks across a wide spectral range, like PET and other polymer substrate materials.For many types of substrates, it is possible to find a spectral region where the sample has peaks but the substrate does not.If the substrate signal interferes with the sample signal, the use of another type of substrate should be considered.Stainless steel, of grade 18/8 or 18/10, is a material that gives close to zero background and is recommended for use as a substrate or sample support, in the case that one can choose freely.A special remark regarding Si substrates is that, in addition to the very strong peak at 520.5 cm À 1 (1 st order), there are weak Si peaks at ca.  970 cm À 1 (2 nd order), ca.1450 cm À 1 (3 rd order), and ca.1930 cm À 1 (4 th order).The 1 st and 2 nd order peaks can be seen in Figure 16a.The 3 rd and 4 th order peaks are easily overlooked and this has resulted in these peaks being attributed to other phenomena. [50]

Dirt and Residue
Most researchers are very careful not to contaminate their samples during synthesis and characterization, but objective lenses with short working distances (< 1 mm) are often used in Raman microscopes and it is easy to unintentionally touch the sample when focusing.To exclude contamination by a previous operator, it is therefore important to check that the objective lens is clean before the measurement is started.
Residues can, of course, be located on other optical components such as on mirrors or filters, but the objective lens is by far the most common location.If signals are coming from a residue on another optical component, these signals will contribute to every spectrum acquired, but could vary in relative intensity compared to the Raman intensity of different samples, and depend on which laser wavelength and intensity is used.

Sample homogeneity, purity, and solvents
All species in the sample are measured, not just the ones that are of interest.It is therefore advised to take some time to consider what is included in the sample and if there can be residues of compounds used in the synthesis or possible byproducts.It is always preferred to have reference spectra or pure reference substances for spectral comparison.Biological samples are very complex and contain many different compounds, including organic molecules, macromolecules, and tissue.In such systems it is difficult to find reference substances and overlap between spectral peaks is virtually unavoidable.Instead, it is common to use a few well defined and unique peaks as markers for larger molecules.For measurement in liquid, it is important to keep in mind that solvents and buffer salts are also compounds with Raman signals.In some research fields an aqueous solution with 0.05 M of a sample compound would be considered highly concentrated but the same solution will also contain 55 M of water molecules that also Raman scatter.As a rule of thumb concentrations below mg/ml in aqueous solutions will be challenging to detect with Raman unless enhancement effects like SERS or resonance Raman are used.The term "detection limit" should be used with caution in Raman spectroscopy.It is not a relevant measure for the Raman technique as a whole and should only be used with clearly specified conditions since it is highly dependent on the scattering of the measured compound.As an extreme, single molecule detection has been proven in several studies using SERS. [51,52]Having a compound dissolved in an organic solvent is even more challenging due to the stronger Raman signal of organic solvents compared to water.

Guidelines for best scientific practice
In the previous sections several tips on how to perform Raman measurements have been presented.These have been focused on how to avoid specific spectral artefacts.
There is also a need to give more general advice and to meet that need we here present guidelines for best scientific practices.A process scheme for Raman measurement is shown in Scheme 1.For a Raman laser, both the laser intensity and emitting wavelength could vary initially so it needs to warm up and stabilize; 15 minutes is usually sufficient.Shorter excitation wavelengths give stronger Raman scattering, as shown in equation 6, but even more important is use of an excitation wavelength that gives low fluorescence in the sample.Before measurements, wipe the objective lenses with optical tissue moistened with ethanol to remove possible contamination.Then focus the laser on a flat surface and view the laser spot, to make sure that the alignment is good.The alignment will affect the calibration, implying that any adjustments to the alignment have to be done before the calibration is checked.Verify that the calibration of the instrument is good.This is usually done by measuring a reference sample with a known, well-defined Raman peak (e.g., Si).Measuring a pair of Stokes/anti-Stokes peaks to verify that they are symmetrical around the Rayleigh line would be considered even more accurate but, as edge filters are more common than notch filters, this might not be an option.Some Raman instruments have a calibration check function inbuilt in the software.Save the measurement as proof that the calibration was correct at the time of the measurement.Note that more than one calibration check might be needed if the system has several laser lines and gratings.Then chose a microscope objective for the measurements.Objectives with high NA, roughly corresponding to high magnification, collect more of the light and give a higher signal.Measure the laser power at the position of the sample, and also consider that the same intensity will be distributed over smaller spot using higher magnification, and thus more likely to induce sample heating effects in the measured spot.There are significant losses in a spectroscopy system and the laser power output decreases over time where, for example, the laser power output is not constant since its manufacture.If the Raman system does not have an enclosure, it is recommended to do one measurement without a sample to see if the system picks up stray light.If reference samples are available it is recommended to start the measurements with them.To have reference spectra available makes it easier to interpret the spectra from the samples that are usually more complex.Reference spectra could include spectra from pure reference samples for compounds suspected to be in the sample, precursors from the synthesis, the solvent, and the substrate.When measuring the sample, start with low laser power and long measurement time.Once the peaks have been identified, increase the laser power stepwise to get better spectrum intensity and signal-to-noise ratio, making sure that the peaks do not change.It is desirable to measure with as high a laser power as possible to reduce the measurement time.Powder samples can be pressed with a spatula, compacting and thinning the powder onto a metal substrate, increasing its heat dissipation away from the measurement spot and enabling the use of higher laser power.
If you encounter fluorescence during measurement that obscures the Raman signal, the recommendation is to switch to a different laser wavelength.For multicomponent samples with micron-sized domains, where the fluorescence originates from a specific compound, it could also be an option to minimize the laser spot (high magnification objective) and use high spectrometer confocality to minimize the probed volume, thus excluding the signal from neighboring fluorescent grains.If your system is equipped with the option for spatial off-set Raman, this can also be used to decrease the PL emission from the surface.Take spectra from several spots on the sample.The measurement spot is usually small in Raman, in the order of magnitude of microns, and to be able to assess whether the acquired spectrum is a good representation of the sample, several measurements are needed.When doing Raman mapping it is highly recommended to first optimize the laser power and find the minimum time needed per spectrum as the total measurement time grows quadratically with the mapped area.When publishing Raman data, it is important that the readers can evaluate the quality of the data and reproduce the measurements.It is therefore recommended to always include the type of instrument or setup that was used, together with the excitation wavelength and the spectral resolution and accuracy in the experimental section.The accuracy depends on both the spectrometer performance and the calibration, and it is therefore recommended to specify that a calibration was done, or checked, in connection with the measurements.We recommend measuring the spectral resolution using an atomic emission line from a calibration light source (e.g., a neon lamp).The spectral resolution depends on the laser wavelength, slit/ pinhole size, grating line density, spectrometer focal length, and the pixel size of the detector, and sometimes these are given instead.When reporting the spectral resolution, it is recommended that it is specified if spectral resolution per pixel or the peak resolution is referred to.In the latter, a minimum of three pixels are needed to determine the peak position and thus, the peak resolution (e.g., a resolution of 0.5 cm À 1 per pixel implies a 1.5 cm À 1 peak resolution).Other parameters that are often relevant to include are the laser power at the sample, the objective magnification and numerical aperture (NA), and measurement time.Specify if any data processing like smoothing or baseline subtraction has been done.

Perspectives
Raman spectroscopy has gone through a significant development over the last 30 years.From being a slow analysis technique mainly used by specialists it has transformed into a convenient everyday tool for scientists in a large variety of disciplines.This transformation has been driven by technological advances in lasers, optical filters, gratings, and detector technology, which has strongly improved the signal throughput and shortened the measurement times by several orders of magnitude.As more non-experts now use Raman, the awareness of the underlying scattering phenomena and its inherent strengths and weaknesses can no longer be taken for granted.In this paper, we draw attention to the fact that Raman scattering is a very weak process, much weaker than several other laser-induced light-emitting processes in materials, and that many external, and internal sources can cause artefacts in the Raman spectra that can be mistaken for Raman peaks from the sample.As a Raman user it is important keep a lookout for abnormal behavior of the peaks.Suspicion should be raised if the peaks are unnaturally sharp or wide, and if they occur randomly or all the time, regardless of sample.Once a false signal has been identified, it is often quite easy to take action.Sometimes it is enough to identify the source and sometimes adjustments to the experiment setup need to be done.It has also been emphasized that it is common to pick up Raman signals from other species in and around the sample and it is therefore important to be aware of the geometry of the setup and the sample composition.By explaining the different artefact-generating phenomena and showing example spectra as well as giving the tools to identify what is causing the artefacts, we hope to provide the tools needed to accurately interpret Raman data, which will save much effort and prevent unnecessary time spent on analyzing signals that are irrelevant to the research.It is our view that difficulties with data interpretation are currently one of the limiting factors of even more widespread use of Raman spectroscopy.The measurement practices presented are one way to avoid many of the artefacts discussed and they can also be used as guidelines for beginners in this field.In the future, Raman spectroscopy is expected to spread into more fields of research and also into other parts of society.This is an ongoing process and applications of Raman are already seen in healthcare diagnostics and customs checks, for example.For such applications where operators are outside the field of science it is desirable that the artefacts described in this paper are avoided or excluded by limitation of the application, instrument design, or by fast-developing software recognition tools rather than by the user.In the field of science, however, the Micro-Raman spectrometers are likely to continue to be the workhorses for many years to come, mainly due to the fact that in research there is a need to measure all types of samples and the flexibility of micro-Raman systems is then needed.

Disclaimer
The opinions expressed in this publication are the views of the authors and do not necessarily reflect the opinions or views of Angewandte Chemie International Edition/Angewandte Chemie, the Publisher, the GDCh, or the affiliated editors.

Figure 1 .
Figure 1.To the left, a Jablonski diagram and Raman spectrum of anatase phase TiO 2 showing both Rayleigh and Stokes and anti-Stokes Raman scattering.Note that the Raman peaks are located symmetrically around the excitation wavelength.The Rayleigh scattered light is strongly attenuated by a blocking filter in this experiment.To the right, principles of advanced variants of Raman, including stimulated Raman, Coherent-Anti Stokes Raman, Resonance Raman, hyper Raman, and surface-and tip-enhanced Raman.

Figure 2 .
Figure 2. a) An aurora borealis in the dark winter sky north of Uppsala, Sweden, caused by solar cosmic rays.b) Raman spectrum of anatase TiO 2 , acquired with a 532 nm laser, with several clearly visible sharp peaks caused by galactic cosmic rays.

Figure 3 .
Figure 3. a) Light shield cylinder with the same inner diameter as the microscope objective's outer diameter.b) Cylinder placed around objective and sample to block out stray light.c) A microscope enclosure can be an integral part of the measurement system or built separately around the microscope.

Figure 4 .
Figure 4. a) The near UV and visible range of a fluorescent light (Aura Luminette) with the dashed square depicting the range shown when using the 532 nm laser.b,c) Peaks from the same fluorescent light as they would appear when measured with the commonly used 532 nm and 785 nm Raman lasers.The data has been baseline subtracted.

Figure 5 .
Figure 5. Spectra of a selection of computer screens: Flatron model W2452TX (upper) and Dell model P2417H (lower).

Figure 6 .
Figure 6.a) Schematic layout of the beam path and optical components in a confocal Raman spectroscopy system.b) Spectrum acquired with a solid-state laser on a non-Raman active metal surface showing both the excitation line (532 nm) and the base frequency of the laser (1064 nm).

Figure 8 .
Figure 8. Raman spectrum showing laser saturation of the detector.The insert shows the low Raman shift region where charge overflow occurs.Data was acquired using a 633 nm laser and a degraded Rayleigh blocking filter.

Figure 9 .
Figure 9. White light profile of a Rayleigh edge filter including a two-step transition and small ripples close to the transition edge.

Figure 10 .
Figure 10.Spectrum acquired from cuprous oxide (Cu 2 O) using a 532 nm laser as excitation source.The Raman signal has been enhanced 5 times to improve its visibility.

Figure 11 .
Figure 11.Laser heating causing phase transitions in iron oxide nanopowder from magnetite to maghemite to hematite.

Figure 12 .
Figure 12.Sample measurement geometries for sample in solution on a glass slide with a) cover glass, and b) thin film sample on sapphire substrate measured in air.The sampling volumes are marked as filled green circles.The sizes of the circles have been exaggerated for visibility reasons and are not to scale.

Figure 14 .
Figure 14.a) Raman spectrum of air showing peaks for oxygen (1555 cm À 1 ) and nitrogen (2329 cm À 1 ) acquired with a 405 nm laser.b) Rotational N 2 and O 2 Raman peaks around the Rayleigh line acquired with a 532 nm laser and a low cut-off Rayleigh notch filter (Eclipse™).

Figure 15 .
Figure 15.Raman spectra of hexagonally close-packed metals Mg, Zn, and Bi that have Raman active vibrations, measured with a 785 nm laser.

Figure 16 .
Figure 16.Raman spectra with peak positions for five common substrates.a) Silicon 514 nm laser excitation, b) sapphire 532 nm laser excitation, c) quartz 532 nm laser excitation, d) fused silica 532 nm laser excitation, e) PET 785 nm laser excitation, and f) 18/8 stainless steel 532 nm laser excitation.The PET spectrum has been baseline subtracted.

Scheme 1 .
Scheme 1. Raman measurement process for suggested scientific best practice.

Table 1 :
A selection of common Raman lasers listed together with their most prominent extra emission lines and attainable wavelengths using higher order harmonics.