Introduction

The exploration of nonequilibrium states of matter using intense and ultrafast light has unveiled previously hidden phases1,2,3,4,5. By employing femtosecond (fs) photoexcitation of electrons in their ground states, highly excited states of potential energy surfaces have been effectively investigated with results revealing previously unknown intermediate phases during relaxation to the ground state or metastable configurations. This approach has promoted the utilization of ultrafast photoinduced dynamics to design material properties on demand through light–matter interactions by controlling the ultrafast energy-relaxation process6,7,8.

A localized surface plasmon (LSP) is excited when light interacts with a metallic nanoparticle smaller than its wavelength. LSPs, with their collective electron oscillations, confine light on the nanoscale to enable precise material control that facilitates active applications in sensing, imaging, energy conversion, and optoelectronics3,9,10,11,12. Coupling LSPs with ultrafast light may control nonequilibrium phase transition kinetics and advance quantum control of materials at nanoscale5,13,14,15. However, research in this field has been limited by the unexamined interactions between LSPs and changing material phases.

In this work, we focus on demonstrating LSP-induced control of ultrafast photoinduced energy-relaxation process by imaging the shape deformation of Au nanorods (NRs) after fs photoexcitation. The Au NRs exhibited significant near-field electromagnetic field enhancement, which was sensitive to particle orientation relative to laser polarization, providing additional control over light–matter interactions. Time-resolved X-ray free-electron laser (XFEL) experiments linked LSP modes to ultrafast photoinduced shape deformation of single Au NRs. Numerical simulations of LSP and acoustic shape distortion modes supported the experimental findings, demonstrating LSP-controlled deformation through different modes in ultrafast energy-relaxation process.

Results

Ultrafast X-ray images show polarization-dependent kinetics

We investigated the influence of LSPs on photoinduced hot electron-initiated shape deformation of single Au NRs accompanying anharmonic lattice deformation. Direct observation of the ultrafast energy-relaxation process was realized using time-resolved single-particle X-ray imaging after exciting the specimens of single Au NRs by a fs infrared (IR) laser pulse (400 nm wavelength with fluence of 300 mJ cm−2 and 800 nm wavelength with 170 mJ cm−2, 230 mJ cm−2 and 350 mJ cm−2, all with 100 fs pulse width). Utilizing the Pohang Accelerator Laboratory X-ray Free-Electron Laser (PAL-XFEL), we collected single-pulse single-particle data for Au NRs mounted on SiNx membranes by raster scanning the membrane relative to the fs-IR laser and X-ray pulse (Fig. 1a and Methods)16,17,18. The random orientation of Au NRs relative to the fs-IR laser polarization enabled the acquisition of single-particle diffraction patterns with laser polarization dependence. We categorized the NRs into three major groups, parallel (Par.), diagonal (Dia.), and perpendicular (Per.), based on their long axis alignment relative to the laser polarization direction (Fig. 1a inset). Different nanoparticle orientations resulted in varying near-field enhancements, leading to different effective electric fields.

Fig. 1: Photoinduced morphological deformation of single Au nanorods (NRs) with polarization-dependent reaction rates.

a Time-resolved single-particle X-ray imaging experiments with single X-ray free electron laser (XFEL) pulses. Au NRs are distributed on SiNx membrane with random orientations to the incident femtosecond (fs)-IR laser polarization. Based on the angle between their long axis and the polarization direction of the incident electric field, Au NRs are categorized into three orientations: Par. (Parallel), Dia. (Diagonal), and Per. (Perpendicular) (inset). b XFEL single-pulse diffraction patterns display morphological changes of the NR from the initial rod to a deformed oval shape. The scale bar is 0.1 nm−1. The purple-to-yellow intensity logarithmic scale indicates low to high photon counts. c The near-field enhancement at the Au NR cross-section is calculated. Color scale represents normalized field enhancement (|E|/|E0|) (dark purple, low; yellow-green, high; linear scale). The scale bar is 50 nm. d Au NRs on photoinduced shape deformation (laser wavelength 800 nm, and the incident fluence at 170 mJ cm−2). The scale bar is 100 nm. The shape deformation speed depends strongly on the laser polarization direction, and the fastest deformation is noted for NRs oriented parallel to the polarization with the highest near-field enhancement. Au NR shape deformation proceeds with local regions of lowered electron density forming from both ends (arrows) of the long axis, progressing inward and forming extended lowered density region along the long axis. Electron density is represented on a blue-to-pink linear scale from low to high. e Absorption spectra of Au NRs on a SiNx substrate for each orientation: Par. (blue), Dia. (orange), and Per. (yellow) orientations. The parallel group exhibits the highest absorption at both 800 nm and 400 nm, consistent with the near-field enhancement pattern, which corresponds to the observed shape deformation speed. Source data are provided as a Source Data file.

Single-pulse XFEL diffraction patterns of single Au NRs photoexcited by a single fs-IR laser pulse (800 nm at the incident laser fluence of 170 mJ cm−2) showed the increase in size, through the energy transfer from photoinduced hot electrons to the lattice, by accompanying a morphological transformation (Fig. 1b)19,20. We also performed a time-resolved wide-angle X-ray diffraction experiment, through the same pump-probe scheme, which revealed the loss of crystalline Bragg reflections after the IR laser pumping, indicating loss of long-range order accompanying the morphological changes (Methods and Supplementary Fig. 1). The lateral size expansion exceeded the longitudinal size to lead to the deformation to an oval shape from the rod. Numerical phase retrieval of all single-pulse diffraction patterns was performed to obtain images displaying the plane-projected electron density of the Au NRs (see “Methods”). Examination of these nanoscale images along three major orientations (Par., Dia., and Per.) revealed significant variation in the deformation speed based on the orientations relative to the fs-IR laser polarization (Fig. 1d and Methods). The fastest and slowest shape deformation occurred for NRs aligned parallel and perpendicular to the laser field, respectively, while maintaining the rod shape until 200 ps. Its deformation speed also depends on the near-field enhancement factor of LSPs (Fig. 1c), obtained from Boundary Element Method (BEM) calculations (Methods). The dependence of the shape deformation speed on the Au NR orientation aligns with the absorption cross-section spectra, which show the highest absorption for Par. and the lowest for Per. (Fig. 1e).

For all three major orientations, the deformation began with a local region of reduced electron density (arrows in Fig. 1d) near the highest electric-field spots of the near-field enhancement. This local density reduction is due to focused plasmonic heating parallel to the laser polarization direction, causing localized ionic pressure accumulation and subsequent structural expansion relieving the accumulated pressure. Similar anisotropic deformation was observed in single-shot imaging experiments on Au nanospheres and in numerical simulations5,6,7,14. The local regions of lower electron density then gradually propagate toward the center along the longitudinal direction, causing the NR to widen along the short axis and eventually developing into an extended region of lowered electron density across the NR. Apart from the reaction speed, the shape deformation process was consistent across all three orientations, with local regions of reduced electron density at both ends of the rod, leading to further lateral expansion and an oval shape.

Deformation rate scales with effective field

Upon its deformation to an oval shape, we quantified the shape deformation rate by tracking changes in the aspect ratio (Fig. 2a and “Methods”). This rate was determined by fitting the temporal evolution of the aspect ratio to an exponential decay function (see “Methods”)21. The shape deformation rate for the parallel orientation was nearly double that of the diagonal orientation and 5 times higher than the perpendicular orientation. This rate is proportional to the effective electric field, indicating that LSPs, excited by the enhanced near-field from fs-IR laser illumination, directly influenced the morphological change. The temporal evolution of the aspect ratio of the Au NRs in all three orientations, normalized by the effective electric field, followed the same fit curve, confirming that LSP is driving this ultrafast energy-relaxation process as a primary energy source (Fig. 2b). Here, we defined the scaled time as ({t}_{{{\rm{scaled}}}}=t\times (\frac{{E}_{{{\rm{eff}}}}}{{E}_{0.}})), where effective electric field (({E}_{{{\rm{eff}}}}(\lambda,\varPhi,\theta ))) is determined by the square root of absorbed energy density divided by vacuum permittivity (Methods). The reference electric field (({E}_{0})) is the effective electric field of the fastest reaction case in the low fluence shape-deformation mode for 800 nm and 170 mJ cm-2 incidence fluence with the orientation of Par. (Fig. 2b and “Methods”).

Fig. 2: Localized surface plasmon (LSP)-induced acoustic shape deformation of the nanorod (NR).

a Shape deformation rates, obtained from the aspect ratio, were extracted and show linear dependence on the effective field. Higher-fluence group shows a steeper slope. b Shape deformation reaction curves (of the lower-fluence group) merge into one exponential decay line with the reaction time linearly scaled for the effective field. a, b Blue, orange, and yellow markers indicate the orientations relative to the laser polarization: Parallel (Par.), diagonal (Dia.), and perpendicular (Per.), respectively. Circles, triangles, diamonds, and squares denote laser fluences and wavelengths of 170 mJ cm−2 (800 nm), 230 mJ cm−2 (800 nm), 300 mJ cm−2 (400 nm), and 350 mJ cm−2 (800 nm), respectively. c Long (blue) and short (yellow) axes of the rods display oscillatory expansion. Black solid lines are fits with the transverse mode frequency of 42.0 GHz (inset). Gray dotted lines indicate frequency shifts. b, c Error bars represent the standard error of the mean. The number of independent nanoparticles at each delay is given in Supplementary Table 6. d Numerical simulation of the LSP mode showing surface charge distribution (red, positive; blue, negative). e Ion displacement during deformation occurs along the transverse direction, with yellow indicating high displacement and black indicating low displacement. f Acoustic deformation mode at 42.0 GHz showing transverse mode. White dotted line indicates the initial undeformed state of the Au NR; color scale (white-to-blue) indicates displacement magnitude. g Volumetric stress field calculated at the acoustic deformation mode. Color scale represents stress magnitude, with red indicating high stress and yellow indicating low stress. d–g Low-absorption mode exhibits transverse shape deformation. Scale bar is 25 nm. Color scales are linear. Au NRs have a length of 145 nm and a diameter of 50 nm. Source data are provided as a Source Data file.

To confirm that the deformation rate was influenced by the effective electric-field strength via LSP excitations, we repeated the experiment for different laser parameters (300 mJ cm−2 at 400 nm and 230 mJ cm−2 at 800 nm). The shape deformation rate, derived from the aspect ratio, maintained a linear relationship with the effective electric field, except for the parallel orientation with a high effective field (1.96 × 102 V nm−1) that showed a different deformation mode (Fig. 2b). Further experiments with an incident laser fluence of 350 mJ cm−2 (at 800 nm) showed similar deformation modes and rate slopes in parallel orientation, aligning with the results obtained at 230 mJ cm−2. We also note that even for the 400 nm excitation, near the resonance energy to excite the transverse LSP, the shape deformation rate was still fastest for the parallel orientation with higher absorption cross section still for the parallel orientation (Supplementary Fig. 7).

Transverse LSP excitation drives transverse deformation

Importantly, the temporal evolution of the long and short axes displayed an oscillatory behavior in addition to the linear expansion, as a feature distinguished from previous studies on nanospheres (Fig. 2c and “Methods”)5,7. This oscillation was characterized by a 42 GHz frequency. As the shape deformation evolved with the aspect ratio decreasing toward an oval shape, the oscillation frequency red-shifted to 35.3 GHz. This suggests that the initial excitation from the intrinsic shape primarily dictates the excited acoustic frequency. The gray dotted line guides this frequency shift behavior caused by an expansion of the NR with the oscillatory deformation (Fig. 2c and “Methods”).

To understand the underlying mechanism of this oscillatory deformation, we performed numerical simulations to figure out LSP-mode excitations, ion displacements, acoustic eigenmodes, and the induced stress field of the NRs (see “Methods”). These simulations replicated different LSP-mode excitations and resulting acoustic deformations. It showed that the oscillatory behavior aligned with transversely driven acoustic shape deformation induced by the transverse plasmon excitation mode (Fig. 2d–g). BEM calculations revealed an Au NR surface plasmon eigenmode, with blue indicating negative and red indicating positive surface charge22,23 (Fig. 2d). We propose that the transverse plasmon mode can be attributed to drive the acoustic deformation with a transverse vibration at 42.0 GHz through plasmon-induced lattice coupling.

Further to verify that the shape deformation of the rod is coupled to the proposed LSP mode, we performed the two-temperature molecular dynamics (TTMD) simulations (Fig. 2e). This TTMD confirmed that the proposed LSP produces local ionic pressure leading to local ionic displacement, with the spatial distribution consistent with the transverse LSP mode, to result in the reported shape deformation (Fig. 2d, e, Supplementary Fig. 9, and “Methods”). Detailed energy transfer process from photoexcited electrons manifested by plasmonic oscillations to lattice was investigated before, which revealed that the surface plasmon with localized high energy foci leads to local accumulation of the ionic pressure launching the local deformation process5,6,7. Finite-element method calculations demonstrated the acoustic distortion and stress field using the eigenmode model with a fixed boundary condition (Fig. 2f, g). The white dotted lines indicate the original shapes of the Au NRs, while the colors represent the normalized displacement and stress fields. The simulations indicate that the transverse deformation mode, driven by the fs-IR laser-excited transverse LSP mode, couples directly to induce the transverse acoustic deformation of the rod. This acoustic deformation, consistent with the reconstructed images, generated a corresponding stress field distribution along the longitudinal direction, resulting in an extended region of lowered electron density across the NR (Fig. 2d–g).

Higher effective field induces longitudinal deformation

In contrast to the deformation behavior observed in the low-fluence group, which involved transverse deformation of the rod and oval shape deformation, a distinct deformation mode was identified in the high-fluence group at 230 mJ cm−2. This process occurred in a parallel orientation, resulting in the formation of two high-density longitudinal loci resembling a dumbbell with one side larger (5 and 8 ps). Initially, the rod elongated as the two loci, marked by arrows, moved apart. Over time, their separation distance decreased (10 ps), resulting in the formation of an elongated oval that approximated the original length but swelled along the short axis (20 ps). Eventually, the shape became rounder with merged high-density spots of expanded size (40 ps). A longitudinal line plot clearly illustrates this process (Fig. 3b and Supplementary Fig. 10). As the morphological deformation of the NR progressed, two high-density regions emerged from 3 ps, diverging from the uniform density of the intact specimen. The distance between these regions increased until (\approx)8 ps, then decreased, leading to the merging of the two high-density regions at (\approx)20 ps. These separation and merging patterns were consistently observed for all the Au NRs at higher effective electric fields (Supplementary Fig. 11).

Fig. 3: Oscillatory longitudinal deformation of Au nanorods (NRs) in a high effective fluence.

a Reconstructed images display the shape deformation of single Au NRs for the parallel orientation in laser fluence of 230 mJ cm−2 with the effective electric field of 1.96 × 102 V nm−1. Deformation proceeds by forming two localized high-density regions along the long axis of the NR, which eventually merge to a circular shape. Electron density is represented on a blue-to-pink scale (low-high). The scale bar is 100 nm. b Oscillating variation of the distance between two high-density loci. Line plot along the long axis displays the process with density change. Green circles indicate data and green lines are guides. c Oscillatory expansion along the long (blue) and short (yellow) axes with the frequency of 51.6 GHz matching with a longitudinal vibration mode. Gray dotted lines represent the frequency shifts. Error bars represent the standard error of the mean. Number of independent nanoparticles at each delay is provided in Supplementary Table 6. d Longitudinal localized surface plasmon (LSP) mode showing surface charge distribution (blue, negative; red, positive). e Ion displacement occurs along the longitudinal plasmon excitation direction. Yellow indicates high displacement and black indicates low. f Acoustic deformation at 51.6 GHz, matching the oscillation frequency of the NRs are obtained from the calculation. White dotted line shows the initial undeformed state; color scale (white-to-blue) indicates displacement magnitude. g Stress field accumulated along the longitudinal axis leads to shape deformation of the NR, with red indicating high stress and yellow low stress. d–g Simulations reproduce longitudinal oscillatory deformation induced by LSP excitation. The scale bar is 25 nm. Au NRs have a length of 145 nm and a diameter of 50 nm. Color scales are linear. Source data are provided as a Source Data file.

Rod expansions along both the long and short axes were derived from all single-particle data (Fig. 3c). Compared with the long axis ((\approx)140%), the expansion along the short axis was significantly larger ((\approx)350%). As aforementioned, the photoinduced shape deformation in the high-fluence group showed an oscillatory pattern alongside linear expansion. This high-fluence energy-relaxation mode, characterized by the structural deformation with longitudinal density separation and merging, is attributed to the photoexcitation of the longitudinal LSP mode, which couples to the lattice to induce the lattice oscillation (Fig. 3d–g). Numerical simulations confirmed that the oscillatory lattice expansion, with a frequency of 51.6 GHz, corresponded to the longitudinal acoustic deformation (Fig. 3f and “Methods”). The higher-fluence deformation mode exhibited an initial oscillation frequency of 51.6 GHz, which becomes red-shifted as the shape deformed into an oval. The gray dotted line is a guideline displaying linearly decreasing frequency from 51.6 to 44 GHz (Fig. 3c and “Methods”). The fs-IR laser field excited the longitudinal surface plasmon mode, causing rod shape distortion with predominant longitudinal deformation. Stress-map estimation indicated local stress confinement along the transverse direction (Fig. 3g), which induced the transient separation and merging of two high-density loci along the longitudinal axis, consistent with experimental observations. The spatial overlap between longitudinal surface plasmon excitation and the lattice induces localized stress, leading to longitudinal shape distortion, as further corroborated by TTMD simulations (Fig. 2d, e, Supplementary Fig. 9, and Methods).

These findings reveal that the nanoscale energy-relaxation process accompanying distinct, fluence dependent, shape-deformation modes can be controlled through the excitation of LSP-coupled anharmonic acoustic deformation. The energy release process from photoexcited hot electrons in Au NRs in the low-absorption case is characterized by the plasma oscillation along the short axis of the rod matching the acoustic deformation eigenfrequency (42.0 GHz), indicating the dominance of the short axis in shape distortion (Fig. 2d–g). In contrast, high-fluence scenarios result in longitudinal shape deformation (Fig. 3), with the oscillation frequency corresponding to the acoustic deformation eigenfrequency (51.6 GHz) induced by the longitudinally vibrating LSP mode. The numerical simulations conducted as part of the study showed that these two energy-relaxation processes with distinct deformation modes are directly caused by anharmonic acoustic deformation mediated by fs-IR-photoinduced LSP modes.

The coupling between LSPs and phonons arises from ultrafast photoexcitation of electrons, resulting in energy transfer to the crystal lattice5,6,7,24,25,26,27. Initially, fs optical excitation of the LSP leads to electron-electron scattering and radiation damping, resulting in an inhomogeneous distribution of hot electrons with a higher average temperature than ions25,28,29,30,31. Equilibration between charge carriers and ions occurs through electron-ion scattering within a few picoseconds after the photoexcitation, triggering lattice deformation from energized ions. Local strain fields are formed in the lattice with the spatial distribution matching with the LSP modes, which is also confirmed by our TTMD simulations (Methods and Supplementary Fig. 9). Such local accumulation of strain, or ionic pressure leads to material deformation starting from hot energy spots formed by LSPs5,6,7. In lower-fluence deformation mode, the LSP mode in the transverse direction is primarily excited, driving shape deformation through strong transverse oscillations. In contrast, higher-fluence deformation mode predominantly excites LSP modes along the longitudinal direction, facilitating deformation through strong longitudinal oscillations. This excitation induces a higher acoustic deformation mode, generating a stress field that deforms the Au NR in the longitudinal direction.

Discussion

This study highlighted the impact of LSP dynamics on the ultrafast energy-relaxation modes of Au NRs with different energy release processes under fs-IR laser excitation. The Au NRs underwent distinct shape deformations influenced by the effective electric field and near-field enhancement. The fs-IR-laser-induced LSP directly controlled the shape-deformation modes through characteristic acoustic lattice deformations. Single-pulse time-resolved imaging at sub-10-nm and picosecond resolutions facilitated direct visualization of the LSP-controlled photoinduced shape-deformation modes, providing evidence of plasmon-ion coupling in light–matter interactions controlling energy-relaxation kinetics. This study deepens the understanding on ultrafast energy transfer process with the potential to manipulate phase changing modes. It advances the control of material kinetics using fs laser pulses and enables quantum control of material properties through ultrafast light–matter interactions. Direct control of phase transition kinetics is enabled employing the LSP mediated mechanical oscillations to suggest a method to custom-design structural properties at the nanoscale. Coupling of nanomaterials with LSP enhances light-matter interactions to promote plasmon-enhanced nano-circuits applications with improved light transport. Further, the nanoscale energy conversion utilizing the LSP excitation offers efficient scheme to devise solar energy harvesting systems.

Methods

Single-pulse single-particle time-resolved X-ray diffraction imaging

Single-pulse single-particle X-ray imaging experiments were performed with a fixed X-ray energy of 5 and 9 keV at the Nanocrystallography and Coherent Imaging (NCI) end station of the PAL-XFEL32. A fs Ti:sapphire laser with a wavelength of 800 and 400 nm, both with pulse duration of 100 fs was used as the pumping source. The laser pulses were focused to a size of 200 μm in root mean square at the position of the samples. A laser linearly polarized in the longitudinal direction was employed to photo-induce the LSPs in the Au NRs. In this study, fs X-ray laser pulses were micron focused using a pair of K-B mirrors with a focal size of 5 μm × 6 μm and a focal length of 5 m. The spatial overlap of the specimens, the 200 μm focused IR laser, and micron-focused XFEL were visually confirmed using an in-line microscope. The time zero of the interaction spot, where all three components (specimens, IR laser, and XFEL) were aligned, was confirmed by monitoring the absorption profile of the thin GaN crystal mounted on the interaction spot. The temporal resolution achieved using the PAL-XFEL was better than 0.5 ps without using a timing tool.

Time-resolved wide-angle X-ray diffraction experiments

Time-resolved wide-angle X-ray diffraction experiments were performed using the focused XFEL radiation (incident X-ray energy of 9 keV) at the NCI station of the PAL-XFEL7,32. The fs-IR laser pumping was performed for 800 nm wavelength with pulse duration of 100 fs. The wide-angle diffraction patterns were collected using the Jungfrau detector installed at 10 cm from the interaction point to track the variation of Au (1 1 1) reflection during photoinduced structural deformation of the Au NRs (Supplementary Fig. 1). Specimens of Au NRs were dispersed on the SiNx membranes.

Fixed-target single particle preparation for XFEL single-pulse data acquisition

Unconjugated rod-shaped Au nanoparticles with a diameter of 50 nm and length of 145 nm, exhibiting peak surface plasmon resonance in the range of 800 nm (NanopartzTM), were mounted on 100 nm thin SiNx membranes. The membranes were custom designed (Silson LTD) with an array of multiple membranes, forming 36 × 2 arrays with individual membrane sizes of 9.0 mm × 0.2 mm, placed on a 25 mm × 25 mm frame. This design facilitated single-pulse diffraction experiments using a fixed-target sample delivery scheme33. To achieve a high particle hit rate, Au NRs in deionized purified water were spin-coated onto SiNx membranes, resulting in a nominal inter-particle distance of 3–5 μm. Single-pulse imaging experiments were conducted, ensuring one XFEL pulse per particle and one IR laser pulse in raster-scan mode. When the SiNx window was exposed to the IR laser pulse, it broke completely. To prevent laser-induced damage to neighboring particles, the shot-to-shot distance was set to 380 μm, greater than the laser footprint of <200 μm. This protocol ensured that each nanoparticle was exposed to a single IR laser and XFEL pulse, while the surrounding SiNx window remained intact. Each single pulse diffraction pattern was recorded on the multiport charge-coupled device detector with a pixel size of 50 μm by 50 μm34. The total detector area with 1024 by 1024 was used.

Coherent diffraction patterns, phase retrieval, and image reconstruction

Coherent single-pulse X-ray diffraction from individual Au nanorods was collected, and the sample was moved to a fresh particle after each exposure5. Diffraction patterns with multi-particle hits, identified by fringe spacing inconsistent with a single object, were discarded (Supplementary Fig. 2). Diffraction intensities were sampled finer than the Nyquist frequency set by the particle size to ensure accurate reconstruction35,36,37.

Real-space images were obtained from each single diffraction pattern by phase retrieval using iterative algorithms16,38. In each reconstruction cycle, the algorithm alternated between reciprocal and real space. The measured diffraction amplitudes were combined with an estimated phase and inverse Fourier transformed to produce a trial image. Real-space constraints were then applied to this image, including a finite support and positivity. The constrained image was Fourier transformed back to reciprocal space, where the calculated amplitudes were replaced with the measured values. Iterations were repeated until the reconstruction error stabilized. To ensure reliability, consistency was verified using multiple random starting phases for each pattern. For each pump-probe delay, a representative image that reflects the average response was selected from several independent reconstructions (Supplementary Movies 1–11).

Workflow for processing and reconstructing diffraction patterns

The diffraction patterns shown in Fig. 1b contained reconstructed pattern as the collected data has center part blocked by the detector. The raw diffraction data we acquired is shown as the “Raw pattern” in Supplementary Fig. 3a. The missing center caused by such beam stopper is retained within the central speckle to reconstruct the images from the measured, center missed data39. The processed pattern is then used to reconstruct the real-space image through iterative phase retrieval algorithms. The “Reconstructed pattern” in the third panel of Supplementary Fig. 3 represents the result after performing an inverse Fourier transform on the retrieved data. This reconstruction is subsequently used to fill the masked regions of the input pattern, yielding the “Filled pattern” shown in the fourth panel. This procedure ensures that the masked regions in the input data are addressed without introducing interpolation or artificial artifacts.

Orientation and aspect ratio determination for each Au NR

We used MATLAB’s regionprops function, which extracts geometric properties from binary images of particles. The centroid (geometric center) was calculated as the arithmetic mean of the pixel coordinates belonging to the particle, representing the average position of all pixels. Using the same function, we also obtained the major axis length (long axis), minor axis length (short axis), and the orientation. Based on this information, we categorized the orientations as Par. (parallel, 0–22.5°), Dia. (diagonal, 22.5°–67.5°), or Per. (perpendicular, 67.5°–112.5°) regarding to the laser polarization direction (Supplementary Fig. 4).

Near-field enhancement and absorption cross-section calculations

Numerical simulations of near-field enhancement for an Au NR on a SiNx substrate under linearly polarized electric fields were performed using BEM, as implemented in the MNPBEM-MATLAB toolbox developed by Hohenester and Trügler22. The BEM simulations were executed in retardation mode, modeling an Au NR with a 50 nm diameter and a total height of 145 nm, positioned above a 100 nm-thick SiNx substrate, matching experimental conditions (Supplementary Fig. 5a). The environment was vacuum, and the mesh discretization used 16, 8, and 15 elements for the azimuthal, polar, and longitudinal directions, respectively, with a total of 420 boundary elements. The complex refractive index of gold was taken from Johnson and Christy40, while SiNx was assigned a refractive index of 2, with the surrounding vacuum set to a refractive index of 1. Simulations were conducted for three different NR orientations, Par., Dia. and Per. by rotating the E-field polarization (0°, 45°, and 90°) while keeping the Au NR orientation fixed. The results include near-field cross-sections on the xy-plane (Fig. 1c), as well as cross-sections of xz-plane and *z-*projected averaged near-field distributions, shown in Supplementary Fig. 5b,c.

For the absorption cross-section spectra, the same simulation setup was used, with calculations performed over a wavelength range of 300 nm to 1200 nm.

Shape deformation rate quantification of the Au NRs

The shape deformation rate was obtained by fitting the temporal evolution of the aspect ratio to an exponential decay function described by the following equation:

$$y={a}_{1} * \exp \left(-\frac{t}{{\tau }_{1}}\right)+{b}_{1}$$

(1)

where (y) is the aspect ratio, (t) is the time delay, ({\tau }_{1}) is the shape deformation time, ({a}_{1}) is the initial amplitude, and ({b}_{1}) is the converging aspect ratio. The fitting results for the shape deformation rate in different laser fluence (170, 230, 300, and 350 mJ cm−2) and orientations of the NR (parallel, diagonal, and perpendicular to the incident laser polarization direction) are summarized in the accompanying table and graph (Supplementary Table 1, Supplementary Fig. 6)

Scaling the shape deformation rate into one curve

We defined the scaled time as ({t}_{{{\rm{scaled}}}}=t\times (\frac{{E}_{{{\rm{eff}}}}}{{E}_{0}})), where effective electric field (({E}_{{{\rm{eff}}}}(\lambda,\varPhi,\theta ))) is given by ({E}_{{{\rm{eff}}}}(\lambda,\varPhi,\theta )=\sqrt{\frac{u}{{{{\rm{\varepsilon }}}}_{0}}}). Here, (u) is the absorbed energy density, determined as (u=\frac{{\partial }_{{{\rm{abs}}}}\left(\lambda,\theta \right)\times F(\lambda,\varPhi,\theta )}{V}), where ({\partial }_{{{\rm{abs}}}}(\lambda,\theta )) is absorption cross-section, (F(\lambda,\varPhi,\theta )) is laser fluence at given (\lambda) (laser wavelength), (\varPhi) (laser fluence), and (\theta) (NR orientation) (see Supplementary Table 2). (V) is the volume of single Au NR with a diameter of 50 nm and a length of 145 nm. The reference electric field (({E}_{0})) is the effective electric field for 800 nm, parallel polarization of 170 mJ cm−2 incidence fluence of the fastest reaction case in the low fluence shape-deformation mode.

To represent the shape deformation rate on a logarithmic scale, the a